Want to use your own AI? Check out our new MCP-only product →
Want to use your own AI? Check out our new MCP-only product →
If you've ever asked your self-hosted OpenClaw a question like "should I rebalance my retirement accounts?" or "am I overexposed to tech?" — you already know what comes back. A neat-sounding paragraph about general investing principles. Maybe a 60/40 allocation example. Definitely no actual numbers. Definitely no idea what you actually own.
That's because OpenClaw — like Claude, ChatGPT, Mistral, and every other large language model — doesn't know your accounts. It has no idea you hold $14,000 in TSLA, no idea your 401(k) is sitting in a 2055 target-date fund at 87 basis points, no idea you have a $3,200 cash drag in your taxable brokerage. It works with the public knowledge it was trained on plus whatever you paste into the chat window.
Truthifi closes that gap. Once you connect Truthifi to your OpenClaw instance, every conversation across every channel — web, Discord, Telegram, custom — has read-only access to your real portfolio. Same OpenClaw you already trust to live on your own server. Same private channels you already use. Now grounded in your actual financial life.
This guide covers what changes when OpenClaw can see your real accounts, what makes a financial MCP connector trustworthy on a self-hosted runtime, and a sample of questions that suddenly become answerable — including the multi-channel scenarios that are uniquely OpenClaw territory.
Why AI gives generic financial answers
Every large language model has the same fundamental limitation when it comes to personal finance: it has no idea who you are or what you own. It can recite the difference between a Roth and a traditional IRA. It can explain dollar-cost averaging. It can describe the trade-offs between active and passive investing. None of that requires knowing anything about you.
The moment you ask a question that depends on your specific situation — "is my asset allocation appropriate for my risk tolerance?" — the model is forced to invent a hypothetical you. It might assume you're 35, earn $120k, have $50k in retirement accounts and $10k in cash. Maybe one of those numbers happens to be right. Most of them won't be. The advice it gives is for the hypothetical you, not the real one.
This is more than an inconvenience. It's the structural reason why "talk to AI about your finances" has been disappointing for everyone who's tried it seriously. The advice you get can only be as specific as the model's understanding of your situation. With no understanding, the advice has no choice but to be generic.
There's a second-order problem too: generic advice that sounds confident is worse than no advice at all. A reader who asks "should I be selling my tech holdings right now?" and gets back a thoughtful-sounding paragraph about not timing the market is being told something that may or may not apply to them. If their tech allocation is already at 80% of their portfolio, the answer should be different than if it's 5%. The agent can't know which case they're in. It picks an answer based on what's plausible for a typical reader and serves that — and a reader who's not typical gets advice tuned for someone else.
For self-hosted OpenClaw users specifically, the friction is even worse. You went out of your way to set up your own AI agent on your own infrastructure — you're not the kind of person who wants to manually paste data five times a day. The whole reason to run OpenClaw was to avoid SaaS dependencies and have an agent that's actually yours. An agent that doesn't know your accounts isn't actually yours in any meaningful way; it's just a model running on your hardware that's no more aware of your situation than the hosted version would be.
Pasting screenshots or CSV exports into the chat helps for one conversation, then becomes stale the moment a position trades. And every paste is a manual chore. The whole reason you set up OpenClaw on your own server was probably to avoid manual chores. Pasting your portfolio into OpenClaw five times a week is the worst possible interaction pattern — friction-heavy, error-prone, and time-bounded by your willingness to keep doing it.
The fix is structural, not procedural: give OpenClaw a live connection to your actual data. Once it's there, you don't need to remember to paste anything. The agent can fetch what it needs when it needs it, mid-conversation, on any channel. The next thirty questions you ask will be grounded automatically. The friction that previously killed the use case disappears.
The two ways to give OpenClaw your financial data
Two paths exist. One is fragile and risky. The other is purpose-built.
The DIY path: scrape, paste, repeat. You log into your brokerage, export a CSV, paste it into OpenClaw. Or you give OpenClaw your brokerage credentials and let it scrape your account portal directly. Both work — kind of — and both have problems. CSVs decay the moment any position trades. Scraping puts your bank credentials on your OpenClaw host, which most users don't want sitting in a Node.js process. And neither approach normalizes data across institutions: Fidelity, Schwab, and Vanguard describe holdings differently, so OpenClaw can't reliably answer "what's my total VTI position across all my accounts?" without you doing the cross-referencing yourself.
There's a deeper issue with the DIY approach that doesn't show up until you've been doing it for a few weeks. Your data and your reasoning drift apart. You ask a question on Tuesday based on Monday's CSV; OpenClaw tells you something. The position changed Monday afternoon. The advice you got on Tuesday is now wrong, but you act on it anyway because you don't realize. This isn't OpenClaw's fault — it's working with what you gave it. But the failure mode is silent, which makes it the worst kind of failure.
For self-hosted OpenClaw, this drift compounds across channels. You might paste a CSV into the web chat on Monday. Tuesday morning your spouse asks the OpenClaw Discord bot a portfolio question — Discord OpenClaw has no idea about Monday's CSV because conversation context is per-channel. You'd have to re-paste. Multi-channel without shared data state means each channel re-introduces the freshness problem.
The purpose-built path: Truthifi's read-only MCP. Truthifi connects to your real financial institutions through industry-standard aggregation, normalizes the data into a consistent schema, and exposes it to OpenClaw via the Model Context Protocol. Your brokerage credentials stay at your brokerage. OpenClaw gets a scoped OAuth token that can read facts and nothing else. When data changes on your accounts, Truthifi reflects it the next time OpenClaw queries.
The MCP layer is the important part. It's an open standard that OpenClaw already supports natively — through the bundled mcporter skill or via the mcpServers config block in ~/.openclaw/openclaw.json. You don't need to build glue code. OpenClaw learns about Truthifi, learns what tools Truthifi exposes, and can call them from any conversation.
The architectural difference matters. With CSV pastes, your data lives in conversation history and ages there. With a scraping tool, your credentials live on your OpenClaw host and become a security obligation. With Truthifi's MCP, the data lives at Truthifi (already a wealth-monitoring service designed to hold it), and OpenClaw queries it on demand. Each component does the thing it's good at: Truthifi does data aggregation and normalization; OpenClaw does conversational reasoning. Neither is doing work it shouldn't.
The multi-channel implication for OpenClaw specifically: once Truthifi is registered at the host level, every channel queries the same data. Web chat sees what Discord sees, and Discord sees what Telegram sees. No re-paste, no per-channel auth, no drift between surfaces.
What makes a financial MCP connector trustworthy
Not all MCP servers are equal. Four things specifically matter for a financial connector that's going to talk to your real accounts.
Read-only by design. Truthifi's MCP server cannot initiate transactions, modify accounts, or move money. There is no write capability anywhere in the connector. This isn't just a permissions configuration — there's no API path through Truthifi to a transactional endpoint at your brokerage. Even if your OpenClaw host were compromised, the worst an attacker could do is read what you can already read.
The asymmetry is structural. When Truthifi originally built its MCP, the team had to decide whether to expose write tools (place trade, transfer money). Some early customers asked for them. The decision to keep the connector read-only was deliberate, and it's encoded in the API surface — not the documentation, the code itself. There is no way to write through it.
Scoped OAuth, not credentials. When OpenClaw connects, Truthifi issues a token that's specifically scoped to the tools OpenClaw needs to call. Your brokerage username and password are never on your OpenClaw host. The token can be revoked from either side at any time.
This matters more on a self-hosted runtime than it does on a SaaS-hosted one. When OpenClaw runs on your own VPS, Raspberry Pi, or AWS instance, you are the one responsible for the security of that host. You're patching it. You're managing access. You're picking the SSH config. A connector that requires you to put your brokerage password on that host turns your host into a credential store. Truthifi's scoped-OAuth model means the worst-case data exposure on a compromised OpenClaw host is the read-only Truthifi token, which is itself revocable in seconds and useless for moving money.
Auditable across channels. OpenClaw's multi-channel design means you can ask portfolio questions from web, Discord, or Telegram. A trustworthy connector logs every tool call with the channel and user identity, so you can audit later — "did anything weird happen on the Discord side last week?" — and answer it. Truthifi maintains a per-user audit log on the Truthifi side; OpenClaw maintains its own log on the host side.
This dual-logging matters because compromise can happen at either layer. If your OpenClaw host is compromised, the Truthifi-side log shows you exactly what an attacker queried, when, and from which channel. If something looks wrong at Truthifi, the OpenClaw-side log tells you whether the call was even initiated by you. Two independent audit trails are the level of due diligence you'd want from any production-grade financial integration.
No data exfiltration. OpenClaw should be the only consumer of your Truthifi token. The connector doesn't ship your data anywhere else, doesn't train on it, doesn't expose it to other Truthifi customers. Your data is yours. Truthifi normalizes it for query, holds it just long enough to answer, and discards working copies.
This is where Truthifi differs from some "AI for finance" SaaS products that exist primarily to harvest user data for model training or competitive intelligence. Truthifi's business model is the data itself — providing accurate, normalized financial data to the user who owns it, via the AI agent of their choice. The user is the customer, not the product.
What to ask once your portfolio is connected
Once Truthifi is registered with OpenClaw, the questions you can productively ask change in shape. Here's a sample, organized by what they reveal — and what a useful answer actually looks like.
Allocation reality check. "Compare my current allocation across all accounts to a 70/25/5 stocks/bonds/cash target. What would I need to do to rebalance?"
A useful response looks like: "Your current allocation is 78% stocks, 17% bonds, 5% cash. To hit 70/25/5, you'd need to sell about $42,000 of stock holdings and buy $42,000 of bonds. The most tax-efficient way to do this is in your tax-advantaged accounts — your 401(k) at Fidelity has $58,000 in equities; rebalancing $42,000 of that to bond funds incurs no tax."
Real numbers, real recommendations, account-specific guidance. Not "consider rebalancing if your allocation has drifted."
Concentration risk. "What's my single largest position as a percentage of total portfolio? And what's my top five concentration?" Useful when you've been holding company stock for a while and aren't sure how much it's grown to dominate.
A useful response: "Your largest position is GOOG, holding $87,200 across your taxable Schwab and IRA accounts — 12.4% of portfolio. Top five (GOOG, AAPL, VTI, VOO, MSFT) total 38.6% of portfolio. Note: GOOG and AAPL are both in tech, so your effective tech concentration through these two is 14.8% of portfolio."
Fee archaeology. "Pull the expense ratios for every mutual fund I hold and rank them. Flag anything over 60 basis points." Most people don't realize they own one or two high-fee funds tucked in a 401(k) menu. OpenClaw can find them in seconds.
A useful response: "Across your 11 mutual fund holdings, the median expense ratio is 22 bps. Three funds exceed 60 bps: AGTHX (Capital Group Growth Fund of America) at 64 bps in your traditional IRA, OAKMX (Oakmark Fund) at 81 bps in your taxable Fidelity, and ABALX (American Funds Balanced) at 56 bps in your 401(k) at Vanguard. Replacing AGTHX with VTSAX would save approximately $186/year on the current $43,000 position. Note: ABALX is below 60 bps but flagged because it's actively managed; VBIAX achieves similar exposure at 8 bps."
Tax-loss harvesting candidates. "Which positions in my taxable accounts are at a loss right now? Show me cost basis and current value, plus the wash-sale window for each." A workflow that used to require a spreadsheet now takes a single question.
A useful response: "You have 3 positions in taxable accounts currently at a loss: BABA at -$2,140 (cost basis $4,800, current $2,660), PYPL at -$890, and DIS at -$340. Wash-sale considerations: the BABA loss can be harvested cleanly (no replacement purchases in past 30 days). PYPL had a small purchase 12 days ago — harvesting now would partially disallow the loss. DIS had no recent purchases, can be harvested freely."
Cross-account totals. "Sum my retirement accounts across all custodians. Break down by account type — Roth, traditional, 401(k), HSA." Especially useful when you've changed jobs and have orphaned accounts at multiple custodians.
Cash drag. "How much cash am I sitting on across all accounts? At current Treasury yields, what's the opportunity cost?" Surfaces money market positions that have crept up over time.
Dividend timing. "What dividends am I expecting in the next 30 days, and from which positions?" Useful for income planning.
Custodian comparison. "Compare my expense ratios at Fidelity to similar funds I could hold at Vanguard. Where would I save the most?" Combines portfolio data with public knowledge — exactly where MCP-grounded LLMs shine.
Multi-channel routine queries. "Show me my net worth change over the past 24 hours" — asked via Discord on a phone in line at the grocery store. Same answer as if you'd asked the OpenClaw web app from your laptop. The multi-channel architecture means the question goes where you are, not the other way around.
Family-channel questions. A shared family Discord with your spouse: "Is our combined portfolio still on track for the retirement target we set at the end of last year?" Both parties can see the answer. Both have visibility. The family-finance conversation that previously required exporting spreadsheets and emailing them happens in the same chat surface where everything else does.
Multi-step workflows. "Pull my five highest-fee funds. Calculate annual fee savings if I moved each to its closest Vanguard equivalent. Estimate the tax cost of the move (assume long-term gains in my taxable, no cost in tax-advantaged). Net it out." OpenClaw chains the queries — fee lookup, equivalent matching, tax estimation, math — and presents a single integrated answer.
The pattern across all of these: OpenClaw doesn't need you to specify which accounts to look at, where the data lives, or what format to use. Truthifi has already normalized everything. OpenClaw queries, reasons, and answers. The cognitive load of "okay let me find the data first" goes away. Multi-channel: the cognitive load of "okay let me re-paste the data on this device" also goes away.
How to connect Truthifi to OpenClaw
The setup takes around five minutes and follows two paths — mcporter skill (recommended, requires Node.js) or direct edit of ~/.openclaw/openclaw.json (works anywhere).
The full step-by-step is here: How to Connect Your Portfolio to OpenClaw via Truthifi MCP.
Short version: install mcporter, run mcporter install --target openclaw truthifi, complete OAuth on first call. Or edit your config to add truthifi to the mcpServers block with type: "http" and url: "https://api.truthifi.com/mcp". Either way, your first portfolio question triggers OAuth, and the connection is live across every OpenClaw channel.
If you run OpenClaw behind a reverse proxy, set MCP_ISSUER_URL to your public HTTPS URL. The setup guide covers why and what happens if you don't.
What OpenClaw can and can't do with your data
A clear list, since this is the question that matters most for trust.
OpenClaw can:
Read your account balances, holdings, cost basis, and transaction history.
Read your Truthifi Score and Truthifi findings (the analytical layer Truthifi runs on top of raw data).
Compute, analyze, summarize, and answer questions across any combination of the above.
Do all of the above from any OpenClaw channel — web, Discord, Telegram, custom — using the same registered Truthifi connection.
Combine portfolio facts with general financial knowledge from its training data — for example, comparing your fund's expense ratio to industry benchmarks, or contextualizing your allocation against age-based recommendations.
Run multi-step workflows that chain Truthifi queries with reasoning — "find the highest-fee fund, find its closest equivalent, compute the savings."
OpenClaw cannot:
Initiate trades, transfers, or any money movement.
Modify your Truthifi or brokerage account settings.
Add, remove, or reauthorize financial-institution connections (you do that in Truthifi).
Access your bank or brokerage credentials. They never leave your bank or brokerage.
Share your data with anyone you haven't shared it with. Truthifi's MCP is yours alone.
Persist your data outside of normal OpenClaw conversation history. Truthifi serves answers; OpenClaw doesn't store full datasets.
Make decisions for you. OpenClaw is good at surfacing facts, doing math, and explaining trade-offs. It is not good at "should I sell" — that's your call.
The asymmetry is intentional: read-everything, write-nothing. It's the right shape for a self-hosted agent. If your OpenClaw host is ever compromised, the read-only constraint means an attacker can't drain your accounts. They can read, which is bad enough — but they can't act.
This is what AI agents were built for
The original promise of AI agents was assistants that could actually help, not just generate plausible-sounding paragraphs. For most use cases — drafting an email, summarizing a meeting, debugging code — that promise has been delivered. For personal finance, it's been mostly stuck, because the agent didn't know your real situation.
OpenClaw plus Truthifi is the missing piece for the self-hosted, privacy-conscious end of that spectrum. You get a real assistant that knows your real portfolio, running on infrastructure you control. No third-party LLM provider is in the loop holding your data. No SaaS dashboard is the source of truth — your own OpenClaw is. Multi-channel, multi-device, all the same data. Discord with your spouse to plan a withdrawal. Telegram to check balance from a hotel. Web to do the deep allocation analysis on your laptop. One connection.
This is also the configuration that holds up best as agents get more capable. As OpenClaw's reasoning improves, as new MCP servers come online for other parts of your financial life (tax software, real estate, debt instruments), the same connector pattern absorbs them. You don't need to redo the OAuth dance each time. You don't need to keep credentials in sync. The MCP layer was designed for exactly this growth.
The longer-term vision is more interesting still. As more financial services expose MCP endpoints — something a number of major institutions are starting to do directly, bypassing aggregation — your OpenClaw becomes a hub for managing your entire financial life conversationally. Tax software exposing return drafts via MCP. Estate-planning tools exposing beneficiary status. Insurance carriers exposing policy details. All of them queryable from the same conversation, the same agent, the same trust model. Your OpenClaw becomes the single chat interface that knows everything you've authorized it to know — and only that.
The reason to set this up now, with Truthifi as the data layer, isn't because you need every one of those questions answered today. It's because the setup is structural. Once your OpenClaw is talking to one MCP server cleanly, adding the next one is trivial. The investment of five minutes today compounds across every future agent improvement.
Multi-channel is going to matter more, not less. Discord agents for family financial conversations, Telegram for quick mobile balance checks, custom voice channels for hands-free queries while driving — same data, same agent, same trust model. The single-channel, single-device assumption that ChatGPT and Claude were built around is starting to feel limiting. OpenClaw was always designed for this multi-channel future. Truthifi makes the data layer fit that design.
There's also a privacy-first angle worth naming. Many users running self-hosted AI made that choice specifically to avoid sending personal data to SaaS LLM providers. With OpenClaw plus Truthifi, the personal financial data flows from Truthifi (where it's already managed under your authority) directly to your OpenClaw host (which you control). The reasoning happens locally on infrastructure you own. The answers stay there. For the privacy-conscious user, this is the configuration that closes the last gap — making AI useful for the most-sensitive personal data without compromising the privacy stance that motivated self-hosting in the first place.
If you've ever asked your self-hosted OpenClaw a question like "should I rebalance my retirement accounts?" or "am I overexposed to tech?" — you already know what comes back. A neat-sounding paragraph about general investing principles. Maybe a 60/40 allocation example. Definitely no actual numbers. Definitely no idea what you actually own.
That's because OpenClaw — like Claude, ChatGPT, Mistral, and every other large language model — doesn't know your accounts. It has no idea you hold $14,000 in TSLA, no idea your 401(k) is sitting in a 2055 target-date fund at 87 basis points, no idea you have a $3,200 cash drag in your taxable brokerage. It works with the public knowledge it was trained on plus whatever you paste into the chat window.
Truthifi closes that gap. Once you connect Truthifi to your OpenClaw instance, every conversation across every channel — web, Discord, Telegram, custom — has read-only access to your real portfolio. Same OpenClaw you already trust to live on your own server. Same private channels you already use. Now grounded in your actual financial life.
This guide covers what changes when OpenClaw can see your real accounts, what makes a financial MCP connector trustworthy on a self-hosted runtime, and a sample of questions that suddenly become answerable — including the multi-channel scenarios that are uniquely OpenClaw territory.
Why AI gives generic financial answers
Every large language model has the same fundamental limitation when it comes to personal finance: it has no idea who you are or what you own. It can recite the difference between a Roth and a traditional IRA. It can explain dollar-cost averaging. It can describe the trade-offs between active and passive investing. None of that requires knowing anything about you.
The moment you ask a question that depends on your specific situation — "is my asset allocation appropriate for my risk tolerance?" — the model is forced to invent a hypothetical you. It might assume you're 35, earn $120k, have $50k in retirement accounts and $10k in cash. Maybe one of those numbers happens to be right. Most of them won't be. The advice it gives is for the hypothetical you, not the real one.
This is more than an inconvenience. It's the structural reason why "talk to AI about your finances" has been disappointing for everyone who's tried it seriously. The advice you get can only be as specific as the model's understanding of your situation. With no understanding, the advice has no choice but to be generic.
There's a second-order problem too: generic advice that sounds confident is worse than no advice at all. A reader who asks "should I be selling my tech holdings right now?" and gets back a thoughtful-sounding paragraph about not timing the market is being told something that may or may not apply to them. If their tech allocation is already at 80% of their portfolio, the answer should be different than if it's 5%. The agent can't know which case they're in. It picks an answer based on what's plausible for a typical reader and serves that — and a reader who's not typical gets advice tuned for someone else.
For self-hosted OpenClaw users specifically, the friction is even worse. You went out of your way to set up your own AI agent on your own infrastructure — you're not the kind of person who wants to manually paste data five times a day. The whole reason to run OpenClaw was to avoid SaaS dependencies and have an agent that's actually yours. An agent that doesn't know your accounts isn't actually yours in any meaningful way; it's just a model running on your hardware that's no more aware of your situation than the hosted version would be.
Pasting screenshots or CSV exports into the chat helps for one conversation, then becomes stale the moment a position trades. And every paste is a manual chore. The whole reason you set up OpenClaw on your own server was probably to avoid manual chores. Pasting your portfolio into OpenClaw five times a week is the worst possible interaction pattern — friction-heavy, error-prone, and time-bounded by your willingness to keep doing it.
The fix is structural, not procedural: give OpenClaw a live connection to your actual data. Once it's there, you don't need to remember to paste anything. The agent can fetch what it needs when it needs it, mid-conversation, on any channel. The next thirty questions you ask will be grounded automatically. The friction that previously killed the use case disappears.
The two ways to give OpenClaw your financial data
Two paths exist. One is fragile and risky. The other is purpose-built.
The DIY path: scrape, paste, repeat. You log into your brokerage, export a CSV, paste it into OpenClaw. Or you give OpenClaw your brokerage credentials and let it scrape your account portal directly. Both work — kind of — and both have problems. CSVs decay the moment any position trades. Scraping puts your bank credentials on your OpenClaw host, which most users don't want sitting in a Node.js process. And neither approach normalizes data across institutions: Fidelity, Schwab, and Vanguard describe holdings differently, so OpenClaw can't reliably answer "what's my total VTI position across all my accounts?" without you doing the cross-referencing yourself.
There's a deeper issue with the DIY approach that doesn't show up until you've been doing it for a few weeks. Your data and your reasoning drift apart. You ask a question on Tuesday based on Monday's CSV; OpenClaw tells you something. The position changed Monday afternoon. The advice you got on Tuesday is now wrong, but you act on it anyway because you don't realize. This isn't OpenClaw's fault — it's working with what you gave it. But the failure mode is silent, which makes it the worst kind of failure.
For self-hosted OpenClaw, this drift compounds across channels. You might paste a CSV into the web chat on Monday. Tuesday morning your spouse asks the OpenClaw Discord bot a portfolio question — Discord OpenClaw has no idea about Monday's CSV because conversation context is per-channel. You'd have to re-paste. Multi-channel without shared data state means each channel re-introduces the freshness problem.
The purpose-built path: Truthifi's read-only MCP. Truthifi connects to your real financial institutions through industry-standard aggregation, normalizes the data into a consistent schema, and exposes it to OpenClaw via the Model Context Protocol. Your brokerage credentials stay at your brokerage. OpenClaw gets a scoped OAuth token that can read facts and nothing else. When data changes on your accounts, Truthifi reflects it the next time OpenClaw queries.
The MCP layer is the important part. It's an open standard that OpenClaw already supports natively — through the bundled mcporter skill or via the mcpServers config block in ~/.openclaw/openclaw.json. You don't need to build glue code. OpenClaw learns about Truthifi, learns what tools Truthifi exposes, and can call them from any conversation.
The architectural difference matters. With CSV pastes, your data lives in conversation history and ages there. With a scraping tool, your credentials live on your OpenClaw host and become a security obligation. With Truthifi's MCP, the data lives at Truthifi (already a wealth-monitoring service designed to hold it), and OpenClaw queries it on demand. Each component does the thing it's good at: Truthifi does data aggregation and normalization; OpenClaw does conversational reasoning. Neither is doing work it shouldn't.
The multi-channel implication for OpenClaw specifically: once Truthifi is registered at the host level, every channel queries the same data. Web chat sees what Discord sees, and Discord sees what Telegram sees. No re-paste, no per-channel auth, no drift between surfaces.
What makes a financial MCP connector trustworthy
Not all MCP servers are equal. Four things specifically matter for a financial connector that's going to talk to your real accounts.
Read-only by design. Truthifi's MCP server cannot initiate transactions, modify accounts, or move money. There is no write capability anywhere in the connector. This isn't just a permissions configuration — there's no API path through Truthifi to a transactional endpoint at your brokerage. Even if your OpenClaw host were compromised, the worst an attacker could do is read what you can already read.
The asymmetry is structural. When Truthifi originally built its MCP, the team had to decide whether to expose write tools (place trade, transfer money). Some early customers asked for them. The decision to keep the connector read-only was deliberate, and it's encoded in the API surface — not the documentation, the code itself. There is no way to write through it.
Scoped OAuth, not credentials. When OpenClaw connects, Truthifi issues a token that's specifically scoped to the tools OpenClaw needs to call. Your brokerage username and password are never on your OpenClaw host. The token can be revoked from either side at any time.
This matters more on a self-hosted runtime than it does on a SaaS-hosted one. When OpenClaw runs on your own VPS, Raspberry Pi, or AWS instance, you are the one responsible for the security of that host. You're patching it. You're managing access. You're picking the SSH config. A connector that requires you to put your brokerage password on that host turns your host into a credential store. Truthifi's scoped-OAuth model means the worst-case data exposure on a compromised OpenClaw host is the read-only Truthifi token, which is itself revocable in seconds and useless for moving money.
Auditable across channels. OpenClaw's multi-channel design means you can ask portfolio questions from web, Discord, or Telegram. A trustworthy connector logs every tool call with the channel and user identity, so you can audit later — "did anything weird happen on the Discord side last week?" — and answer it. Truthifi maintains a per-user audit log on the Truthifi side; OpenClaw maintains its own log on the host side.
This dual-logging matters because compromise can happen at either layer. If your OpenClaw host is compromised, the Truthifi-side log shows you exactly what an attacker queried, when, and from which channel. If something looks wrong at Truthifi, the OpenClaw-side log tells you whether the call was even initiated by you. Two independent audit trails are the level of due diligence you'd want from any production-grade financial integration.
No data exfiltration. OpenClaw should be the only consumer of your Truthifi token. The connector doesn't ship your data anywhere else, doesn't train on it, doesn't expose it to other Truthifi customers. Your data is yours. Truthifi normalizes it for query, holds it just long enough to answer, and discards working copies.
This is where Truthifi differs from some "AI for finance" SaaS products that exist primarily to harvest user data for model training or competitive intelligence. Truthifi's business model is the data itself — providing accurate, normalized financial data to the user who owns it, via the AI agent of their choice. The user is the customer, not the product.
What to ask once your portfolio is connected
Once Truthifi is registered with OpenClaw, the questions you can productively ask change in shape. Here's a sample, organized by what they reveal — and what a useful answer actually looks like.
Allocation reality check. "Compare my current allocation across all accounts to a 70/25/5 stocks/bonds/cash target. What would I need to do to rebalance?"
A useful response looks like: "Your current allocation is 78% stocks, 17% bonds, 5% cash. To hit 70/25/5, you'd need to sell about $42,000 of stock holdings and buy $42,000 of bonds. The most tax-efficient way to do this is in your tax-advantaged accounts — your 401(k) at Fidelity has $58,000 in equities; rebalancing $42,000 of that to bond funds incurs no tax."
Real numbers, real recommendations, account-specific guidance. Not "consider rebalancing if your allocation has drifted."
Concentration risk. "What's my single largest position as a percentage of total portfolio? And what's my top five concentration?" Useful when you've been holding company stock for a while and aren't sure how much it's grown to dominate.
A useful response: "Your largest position is GOOG, holding $87,200 across your taxable Schwab and IRA accounts — 12.4% of portfolio. Top five (GOOG, AAPL, VTI, VOO, MSFT) total 38.6% of portfolio. Note: GOOG and AAPL are both in tech, so your effective tech concentration through these two is 14.8% of portfolio."
Fee archaeology. "Pull the expense ratios for every mutual fund I hold and rank them. Flag anything over 60 basis points." Most people don't realize they own one or two high-fee funds tucked in a 401(k) menu. OpenClaw can find them in seconds.
A useful response: "Across your 11 mutual fund holdings, the median expense ratio is 22 bps. Three funds exceed 60 bps: AGTHX (Capital Group Growth Fund of America) at 64 bps in your traditional IRA, OAKMX (Oakmark Fund) at 81 bps in your taxable Fidelity, and ABALX (American Funds Balanced) at 56 bps in your 401(k) at Vanguard. Replacing AGTHX with VTSAX would save approximately $186/year on the current $43,000 position. Note: ABALX is below 60 bps but flagged because it's actively managed; VBIAX achieves similar exposure at 8 bps."
Tax-loss harvesting candidates. "Which positions in my taxable accounts are at a loss right now? Show me cost basis and current value, plus the wash-sale window for each." A workflow that used to require a spreadsheet now takes a single question.
A useful response: "You have 3 positions in taxable accounts currently at a loss: BABA at -$2,140 (cost basis $4,800, current $2,660), PYPL at -$890, and DIS at -$340. Wash-sale considerations: the BABA loss can be harvested cleanly (no replacement purchases in past 30 days). PYPL had a small purchase 12 days ago — harvesting now would partially disallow the loss. DIS had no recent purchases, can be harvested freely."
Cross-account totals. "Sum my retirement accounts across all custodians. Break down by account type — Roth, traditional, 401(k), HSA." Especially useful when you've changed jobs and have orphaned accounts at multiple custodians.
Cash drag. "How much cash am I sitting on across all accounts? At current Treasury yields, what's the opportunity cost?" Surfaces money market positions that have crept up over time.
Dividend timing. "What dividends am I expecting in the next 30 days, and from which positions?" Useful for income planning.
Custodian comparison. "Compare my expense ratios at Fidelity to similar funds I could hold at Vanguard. Where would I save the most?" Combines portfolio data with public knowledge — exactly where MCP-grounded LLMs shine.
Multi-channel routine queries. "Show me my net worth change over the past 24 hours" — asked via Discord on a phone in line at the grocery store. Same answer as if you'd asked the OpenClaw web app from your laptop. The multi-channel architecture means the question goes where you are, not the other way around.
Family-channel questions. A shared family Discord with your spouse: "Is our combined portfolio still on track for the retirement target we set at the end of last year?" Both parties can see the answer. Both have visibility. The family-finance conversation that previously required exporting spreadsheets and emailing them happens in the same chat surface where everything else does.
Multi-step workflows. "Pull my five highest-fee funds. Calculate annual fee savings if I moved each to its closest Vanguard equivalent. Estimate the tax cost of the move (assume long-term gains in my taxable, no cost in tax-advantaged). Net it out." OpenClaw chains the queries — fee lookup, equivalent matching, tax estimation, math — and presents a single integrated answer.
The pattern across all of these: OpenClaw doesn't need you to specify which accounts to look at, where the data lives, or what format to use. Truthifi has already normalized everything. OpenClaw queries, reasons, and answers. The cognitive load of "okay let me find the data first" goes away. Multi-channel: the cognitive load of "okay let me re-paste the data on this device" also goes away.
How to connect Truthifi to OpenClaw
The setup takes around five minutes and follows two paths — mcporter skill (recommended, requires Node.js) or direct edit of ~/.openclaw/openclaw.json (works anywhere).
The full step-by-step is here: How to Connect Your Portfolio to OpenClaw via Truthifi MCP.
Short version: install mcporter, run mcporter install --target openclaw truthifi, complete OAuth on first call. Or edit your config to add truthifi to the mcpServers block with type: "http" and url: "https://api.truthifi.com/mcp". Either way, your first portfolio question triggers OAuth, and the connection is live across every OpenClaw channel.
If you run OpenClaw behind a reverse proxy, set MCP_ISSUER_URL to your public HTTPS URL. The setup guide covers why and what happens if you don't.
What OpenClaw can and can't do with your data
A clear list, since this is the question that matters most for trust.
OpenClaw can:
Read your account balances, holdings, cost basis, and transaction history.
Read your Truthifi Score and Truthifi findings (the analytical layer Truthifi runs on top of raw data).
Compute, analyze, summarize, and answer questions across any combination of the above.
Do all of the above from any OpenClaw channel — web, Discord, Telegram, custom — using the same registered Truthifi connection.
Combine portfolio facts with general financial knowledge from its training data — for example, comparing your fund's expense ratio to industry benchmarks, or contextualizing your allocation against age-based recommendations.
Run multi-step workflows that chain Truthifi queries with reasoning — "find the highest-fee fund, find its closest equivalent, compute the savings."
OpenClaw cannot:
Initiate trades, transfers, or any money movement.
Modify your Truthifi or brokerage account settings.
Add, remove, or reauthorize financial-institution connections (you do that in Truthifi).
Access your bank or brokerage credentials. They never leave your bank or brokerage.
Share your data with anyone you haven't shared it with. Truthifi's MCP is yours alone.
Persist your data outside of normal OpenClaw conversation history. Truthifi serves answers; OpenClaw doesn't store full datasets.
Make decisions for you. OpenClaw is good at surfacing facts, doing math, and explaining trade-offs. It is not good at "should I sell" — that's your call.
The asymmetry is intentional: read-everything, write-nothing. It's the right shape for a self-hosted agent. If your OpenClaw host is ever compromised, the read-only constraint means an attacker can't drain your accounts. They can read, which is bad enough — but they can't act.
This is what AI agents were built for
The original promise of AI agents was assistants that could actually help, not just generate plausible-sounding paragraphs. For most use cases — drafting an email, summarizing a meeting, debugging code — that promise has been delivered. For personal finance, it's been mostly stuck, because the agent didn't know your real situation.
OpenClaw plus Truthifi is the missing piece for the self-hosted, privacy-conscious end of that spectrum. You get a real assistant that knows your real portfolio, running on infrastructure you control. No third-party LLM provider is in the loop holding your data. No SaaS dashboard is the source of truth — your own OpenClaw is. Multi-channel, multi-device, all the same data. Discord with your spouse to plan a withdrawal. Telegram to check balance from a hotel. Web to do the deep allocation analysis on your laptop. One connection.
This is also the configuration that holds up best as agents get more capable. As OpenClaw's reasoning improves, as new MCP servers come online for other parts of your financial life (tax software, real estate, debt instruments), the same connector pattern absorbs them. You don't need to redo the OAuth dance each time. You don't need to keep credentials in sync. The MCP layer was designed for exactly this growth.
The longer-term vision is more interesting still. As more financial services expose MCP endpoints — something a number of major institutions are starting to do directly, bypassing aggregation — your OpenClaw becomes a hub for managing your entire financial life conversationally. Tax software exposing return drafts via MCP. Estate-planning tools exposing beneficiary status. Insurance carriers exposing policy details. All of them queryable from the same conversation, the same agent, the same trust model. Your OpenClaw becomes the single chat interface that knows everything you've authorized it to know — and only that.
The reason to set this up now, with Truthifi as the data layer, isn't because you need every one of those questions answered today. It's because the setup is structural. Once your OpenClaw is talking to one MCP server cleanly, adding the next one is trivial. The investment of five minutes today compounds across every future agent improvement.
Multi-channel is going to matter more, not less. Discord agents for family financial conversations, Telegram for quick mobile balance checks, custom voice channels for hands-free queries while driving — same data, same agent, same trust model. The single-channel, single-device assumption that ChatGPT and Claude were built around is starting to feel limiting. OpenClaw was always designed for this multi-channel future. Truthifi makes the data layer fit that design.
There's also a privacy-first angle worth naming. Many users running self-hosted AI made that choice specifically to avoid sending personal data to SaaS LLM providers. With OpenClaw plus Truthifi, the personal financial data flows from Truthifi (where it's already managed under your authority) directly to your OpenClaw host (which you control). The reasoning happens locally on infrastructure you own. The answers stay there. For the privacy-conscious user, this is the configuration that closes the last gap — making AI useful for the most-sensitive personal data without compromising the privacy stance that motivated self-hosting in the first place.
The smartest money move you can make? Hook it up to AI.
Truthifi® tests your finances for 100+ risks and opportunities—automatically. Unlock plain-English insights that drive smarter financial decisions today.
The smartest money move you can make? Hook it up to AI.
Truthifi® tests your finances for 100+ risks and opportunities—automatically. Unlock plain-English insights that drive smarter financial decisions today.
The smartest money move you can make? Hook it up to AI.
Truthifi® tests your finances for 100+ risks and opportunities—automatically.
Frequently asked questions
Do I need a paid Truthifi plan to use this? A free Truthifi account is enough to connect one financial institution. Pro and Max accounts add unlimited institutions and the analytical layer (Truthifi Score, findings, fee analysis). The connection mechanics are identical regardless of plan.
Can multiple people share an OpenClaw host with separate Truthifi accounts? Yes. Each user authenticates separately the first time they ask a Truthifi-touching prompt. Tokens are stored per-user on the OpenClaw host. There's no cross-user data leakage — user A's queries return user A's portfolio, full stop.
What happens to my OpenClaw conversation history? OpenClaw stores conversation history locally on your host. The Truthifi data that flows through those conversations stays in conversation history per OpenClaw's normal retention rules. Truthifi itself doesn't retain copies of your conversations.
How quickly does Truthifi reflect changes from my brokerage? Truthifi syncs at standard aggregation intervals — typically once per business day for most institutions, real-time or near-real-time for some. When you ask OpenClaw about a position that just traded, you'll see the most recent sync, which is usually less than 24 hours stale.
What if I want to add a different MCP server alongside Truthifi? That's the design point of OpenClaw's mcpServers config block. Add as many servers as you like — Truthifi for portfolio data, GitHub for code, Notion for notes, anything else MCP-compatible. They coexist; OpenClaw routes queries to whichever server has the right tool.
Is the connection encrypted end-to-end? Yes. OpenClaw talks to Truthifi over HTTPS with TLS. OAuth tokens are stored encrypted on your OpenClaw host. Truthifi's connection to your underlying brokerages goes over their respective secure APIs.
Can I see what tools Truthifi exposes to OpenClaw? Yes. From OpenClaw, run mcporter list truthifi (Path A users) or check the OpenClaw gateway logs for the tool advertisement on first call. Truthifi also publishes the tool schema at https://api.truthifi.com/mcp — OpenClaw fetches this on connection.
What if OpenClaw gives me an answer that contradicts what I see in my brokerage app? Almost always a sync timing issue. Truthifi reflects what the brokerage exposed at last sync; if a trade just settled, Truthifi may not have it yet. Re-ask in a few hours. If discrepancies persist, check the audit log at truthifi.com to confirm which sync the answer was based on.
Does OpenClaw's reasoning ever leak my data through generic outputs? OpenClaw's reasoning happens locally on your host. The model itself doesn't transmit your data to a third-party provider. If you've configured OpenClaw to route to a hosted LLM (some users do), the data flows through that provider. Most OpenClaw users running fully self-hosted don't have this concern.
Can I use Truthifi from a Discord bot specifically? Yes. OpenClaw's Discord channel uses the same registered Truthifi connector as the web channel. Bot commands, slash commands, or natural-language queries all work the same way. Useful for shared family-finance servers where multiple people want quick access to portfolio facts.
What about voice — can OpenClaw answer portfolio questions verbally? If you've extended OpenClaw with a voice channel (some users do via custom integrations), Truthifi data is available there same as anywhere else. Speak the question, hear the answer, real portfolio data underneath. The MCP layer doesn't care what surface the conversation is on.
Stop asking OpenClaw about a hypothetical portfolio
Every minute you spend pasting screenshots into your AI agent is a minute it's working with stale data. Every "could you analyze my situation" prompt that ends with "I don't have access to your accounts" is a friction point that shouldn't exist. The whole reason you run OpenClaw on your own server is to have a real assistant under your control. Give it real data.
Connect Truthifi → ask OpenClaw the question that's been on your list. The setup takes five minutes. The first useful answer pays it back immediately.
Setup guide: How to Connect Your Portfolio to OpenClaw via Truthifi MCP.
Frequently asked questions
Do I need a paid Truthifi plan to use this? A free Truthifi account is enough to connect one financial institution. Pro and Max accounts add unlimited institutions and the analytical layer (Truthifi Score, findings, fee analysis). The connection mechanics are identical regardless of plan.
Can multiple people share an OpenClaw host with separate Truthifi accounts? Yes. Each user authenticates separately the first time they ask a Truthifi-touching prompt. Tokens are stored per-user on the OpenClaw host. There's no cross-user data leakage — user A's queries return user A's portfolio, full stop.
What happens to my OpenClaw conversation history? OpenClaw stores conversation history locally on your host. The Truthifi data that flows through those conversations stays in conversation history per OpenClaw's normal retention rules. Truthifi itself doesn't retain copies of your conversations.
How quickly does Truthifi reflect changes from my brokerage? Truthifi syncs at standard aggregation intervals — typically once per business day for most institutions, real-time or near-real-time for some. When you ask OpenClaw about a position that just traded, you'll see the most recent sync, which is usually less than 24 hours stale.
What if I want to add a different MCP server alongside Truthifi? That's the design point of OpenClaw's mcpServers config block. Add as many servers as you like — Truthifi for portfolio data, GitHub for code, Notion for notes, anything else MCP-compatible. They coexist; OpenClaw routes queries to whichever server has the right tool.
Is the connection encrypted end-to-end? Yes. OpenClaw talks to Truthifi over HTTPS with TLS. OAuth tokens are stored encrypted on your OpenClaw host. Truthifi's connection to your underlying brokerages goes over their respective secure APIs.
Can I see what tools Truthifi exposes to OpenClaw? Yes. From OpenClaw, run mcporter list truthifi (Path A users) or check the OpenClaw gateway logs for the tool advertisement on first call. Truthifi also publishes the tool schema at https://api.truthifi.com/mcp — OpenClaw fetches this on connection.
What if OpenClaw gives me an answer that contradicts what I see in my brokerage app? Almost always a sync timing issue. Truthifi reflects what the brokerage exposed at last sync; if a trade just settled, Truthifi may not have it yet. Re-ask in a few hours. If discrepancies persist, check the audit log at truthifi.com to confirm which sync the answer was based on.
Does OpenClaw's reasoning ever leak my data through generic outputs? OpenClaw's reasoning happens locally on your host. The model itself doesn't transmit your data to a third-party provider. If you've configured OpenClaw to route to a hosted LLM (some users do), the data flows through that provider. Most OpenClaw users running fully self-hosted don't have this concern.
Can I use Truthifi from a Discord bot specifically? Yes. OpenClaw's Discord channel uses the same registered Truthifi connector as the web channel. Bot commands, slash commands, or natural-language queries all work the same way. Useful for shared family-finance servers where multiple people want quick access to portfolio facts.
What about voice — can OpenClaw answer portfolio questions verbally? If you've extended OpenClaw with a voice channel (some users do via custom integrations), Truthifi data is available there same as anywhere else. Speak the question, hear the answer, real portfolio data underneath. The MCP layer doesn't care what surface the conversation is on.
Stop asking OpenClaw about a hypothetical portfolio
Every minute you spend pasting screenshots into your AI agent is a minute it's working with stale data. Every "could you analyze my situation" prompt that ends with "I don't have access to your accounts" is a friction point that shouldn't exist. The whole reason you run OpenClaw on your own server is to have a real assistant under your control. Give it real data.
Connect Truthifi → ask OpenClaw the question that's been on your list. The setup takes five minutes. The first useful answer pays it back immediately.
Setup guide: How to Connect Your Portfolio to OpenClaw via Truthifi MCP.
Disclaimer: This article is for educational purposes only and does not constitute financial, tax, or legal advice. It should not be construed as a personalized recommendation regarding any investment, financial advisor, or financial product. All calculations use hypothetical scenarios and historical return assumptions; actual results will vary. Past performance does not guarantee future results. Consult a qualified financial professional for guidance specific to your situation. Truthifi is an investment monitoring platform — not a financial advisor, broker-dealer, or tax professional. Truthifi does not manage assets, recommend investments, sell financial products, or provide personalized financial advice. Truthifi earns no revenue from advisor referrals, product commissions, or AUM fees. Statistics and data cited reflect publicly available sources current as of the article's publication date. Sources are linked throughout.
Disclaimer: This article is for educational purposes only and does not constitute financial, tax, or legal advice. It should not be construed as a personalized recommendation regarding any investment, financial advisor, or financial product. All calculations use hypothetical scenarios and historical return assumptions; actual results will vary. Past performance does not guarantee future results. Consult a qualified financial professional for guidance specific to your situation. Truthifi is an investment monitoring platform — not a financial advisor, broker-dealer, or tax professional. Truthifi does not manage assets, recommend investments, sell financial products, or provide personalized financial advice. Truthifi earns no revenue from advisor referrals, product commissions, or AUM fees. Statistics and data cited reflect publicly available sources current as of the article's publication date. Sources are linked throughout.
Disclaimer: This article is for educational purposes only and does not constitute financial, tax, or legal advice. It should not be construed as a personalized recommendation regarding any investment, financial advisor, or financial product. All calculations use hypothetical scenarios and historical return assumptions; actual results will vary. Past performance does not guarantee future results. Consult a qualified financial professional for guidance specific to your situation. Truthifi is an investment monitoring platform — not a financial advisor, broker-dealer, or tax professional. Truthifi does not manage assets, recommend investments, sell financial products, or provide personalized financial advice. Truthifi earns no revenue from advisor referrals, product commissions, or AUM fees. Statistics and data cited reflect publicly available sources current as of the article's publication date. Sources are linked throughout.
Ready to get started?