Sure MCP Server
Self-hosted MCP endpoint built into Sure Finance for external AI assistants that need token-authenticated access to personal finance data, accounts, transactions, holdings, budgets, balance sheets, income statements, uploaded family files, bank-statement imports, and goal creation.
Open the source and read safety notes before installing.
Safety notes
- Sure MCP exposes personal finance tools over a token-authenticated endpoint. Treat the bearer token as access to the configured user's family finance data.
- Current code accepts Doorkeeper bearer tokens with `read_write` scope and also supports `MCP_API_TOKEN` plus `MCP_USER_EMAIL` as an env-token fallback.
- Most tools read financial data, but `create_goal` creates a Sure goal and `import_bank_statement` can create a transaction import from an uploaded bank-statement PDF.
- The AI compose file recommends routing external AI traffic through Pipelock's MCP reverse proxy on port 8889, but it also notes that `/mcp` remains reachable on the web port when published.
- The source version checked during review was `0.7.2-alpha.7`, while the latest GitHub release was `v0.7.1-hotfix.1`; verify deployment maturity before relying on it for production financial workflows.
- Do not connect untrusted MCP clients, broad autonomous agents, or public endpoints to a Sure instance containing real account, transaction, investment, tax, or document data.
Privacy notes
- MCP tools can expose account names, balances, linked-provider status, transaction history, merchants, categories, tags, holdings, securities, budgets, income, expenses, net worth, goals, and family file search results.
- The bank-statement import tool can send uploaded PDF content to the configured LLM provider for extraction and then create an import for review.
- The family-file search tool can expose excerpts from uploaded documents such as tax returns, bank statements, contracts, insurance policies, investment reports, CSVs, spreadsheets, and other stored files.
- OpenAI-compatible providers, Anthropic support, external assistants, OpenClaw, Pipelock, vector stores, Langfuse tracing, proxy logs, Rails logs, OAuth tokens, and bearer headers may all become part of the data-flow boundary depending on configuration.
- Keep `MCP_API_TOKEN`, OAuth access tokens, `EXTERNAL_ASSISTANT_TOKEN`, provider keys, document excerpts, transaction exports, and Pipelock logs out of public prompts, issues, screenshots, and commits.
Prerequisites
- A self-hosted Sure instance or a Sure deployment you administer.
- Docker Compose or another supported Sure deployment path, plus PostgreSQL, Redis, storage, and web access configured for the app.
- `MCP_API_TOKEN` and `MCP_USER_EMAIL` for env-token auth, or an OAuth/Doorkeeper access token with `read_write` scope.
- A Sure user whose email matches `MCP_USER_EMAIL` when using env-token auth; the MCP server accesses that user's family finance data.
- An MCP or JSON-RPC client that supports a remote HTTP endpoint with bearer-token headers, or a proxy layer that adapts the endpoint for your client.
- A privacy and approval plan for financial data, uploaded documents, goal creation, bank-statement import, logs, traces, LLM providers, and external assistant routing.
Schema details
- Install type
- cli
- Troubleshooting
- No
- Scope
- Source repo
- Estimated setup
- 45 minutes
- Difficulty
- advanced
- Website
- https://sure.am
- Pricing
- open-source
- Disclosure
- editorial
- Application category
- FinanceApplication
- Operating system
- Cross-platform
Full copyable content
curl -o compose.yml https://raw.githubusercontent.com/we-promise/sure/main/compose.example.yml
curl -o compose.ai.yml https://raw.githubusercontent.com/we-promise/sure/main/compose.example.ai.yml
curl -o pipelock.example.yaml https://raw.githubusercontent.com/we-promise/sure/main/pipelock.example.yaml
# .env
MCP_API_TOKEN=generate-a-random-token-here
MCP_USER_EMAIL=user@example.com
docker compose -f compose.ai.yml up -dAbout this resource
Overview
Sure MCP Server is the remote MCP endpoint built into
we-promise/sure, the community-maintained Sure Finance fork of Maybe Finance.
It lets external AI assistants query a self-hosted Sure instance through
JSON-RPC 2.0 methods for initialization, tool listing, and tool calls.
Use it when a finance-aware assistant needs structured access to a user's Sure data instead of copy-pasted exports. The strongest fit is a self-hosted Sure deployment where the operator wants Claude, OpenClaw, GPT agents, or a custom agent to analyze accounts, transactions, budgets, holdings, documents, and goals while keeping Sure as the system of record.
Status
Verified on 2026-06-18:
- GitHub repository:
we-promise/sure - Current source version in
.sure-version:0.7.2-alpha.7 - Latest GitHub release found during review:
v0.7.1-hotfix.1, published on 2026-06-11 - License: AGPL-3.0
- GitHub metadata showed more than 8,700 stars and activity on 2026-06-18
- MCP implementation is present in
main, documented indocs/hosting/mcp.md, routed asPOST /mcp, exposed through Settings > MCP, and covered by controller tests
This entry treats the MCP surface as real but still early-maturity. Re-check the
upstream docs, release notes, and .sure-version before using it for production
financial workflows.
Install
Sure's self-hosting docs use Docker Compose as the recommended deployment path. For the AI and MCP stack, fetch the standard compose file, the AI compose file, and the Pipelock example configuration:
curl -o compose.yml https://raw.githubusercontent.com/we-promise/sure/main/compose.example.yml
curl -o compose.ai.yml https://raw.githubusercontent.com/we-promise/sure/main/compose.example.ai.yml
curl -o pipelock.example.yaml https://raw.githubusercontent.com/we-promise/sure/main/pipelock.example.yaml
Set MCP credentials in .env:
MCP_API_TOKEN=generate-a-random-token-here
MCP_USER_EMAIL=user@example.com
Then start the AI compose stack:
docker compose -f compose.ai.yml up -d
The MCP docs describe direct access at POST /mcp. The AI compose file also
configures Pipelock as an MCP reverse proxy on port 8889 and recommends
external AI clients connect through that proxy for scanning.
Client Configuration
For remote MCP clients that support HTTP endpoints and headers, connect directly to Sure:
{
"mcpServers": {
"sure": {
"url": "https://your-sure-instance/mcp",
"headers": {
"Authorization": "Bearer ${SURE_MCP_TOKEN}"
}
}
}
}
For deployments using Pipelock, point the client at the Pipelock MCP proxy instead:
https://your-pipelock-proxy.example.com
Some stdio-only MCP clients may need a bridge or proxy because Sure's endpoint is a remote JSON-RPC HTTP endpoint rather than a local stdio server.
Protocol and Auth
The McpController implements:
| Method | Purpose |
|---|---|
initialize |
Returns protocol version 2025-03-26, tool capability metadata, and server info |
tools/list |
Returns Sure assistant function tools with names, descriptions, and input schemas |
tools/call |
Executes one Sure assistant function and returns JSON as MCP text content |
Authentication paths:
| Path | Source Behavior |
|---|---|
| Doorkeeper access token | Current controller accepts bearer tokens for active users when token scope includes read_write |
| Env-token fallback | MCP_API_TOKEN must match the bearer token and MCP_USER_EMAIL must match an existing Sure user |
| OAuth metadata | .well-known metadata advertises authorization and token endpoints, dynamic registration, PKCE S256, and read_write scope |
The settings page displays the instance MCP URL and lets a user revoke connected Doorkeeper tokens.
Tool Scope
The current Assistant.function_classes registry includes:
| Tool | Scope |
|---|---|
get_transactions |
Search and page through transactions by date, account, category, merchant, tag, amount, and order |
get_accounts |
Return account names, balances, classification, type, provider, status, and historical balances |
get_holdings |
Search current investment and crypto holdings with account and security filters |
get_balance_sheet |
Return assets, liabilities, net worth, monthly history, and debt-to-asset insight |
get_income_statement |
Aggregate income, expenses, categories, net income, savings rate, and monthly spending metrics |
get_budget |
Show monthly budget progress and prior-month comparison data |
import_bank_statement |
Extract transactions from an uploaded bank-statement PDF and create a transaction import for review |
search_family_files |
Search uploaded family documents through the configured vector store |
create_goal |
Create a family savings goal linked to depository accounts after confirmation |
Best Use Cases
- Ask an external assistant to explain spending patterns from Sure transactions without exporting CSVs into a chat.
- Build a local OpenClaw or custom agent that calls Sure for finance context and keeps conversational orchestration outside the Rails app.
- Query account balances, investment holdings, net worth, income statements, and budget trends through a structured tool interface.
- Search uploaded financial documents, tax files, insurance policies, or bank statements from a family vault.
- Route inbound MCP traffic through Pipelock for DLP, prompt-injection, tool poisoning, and tool-call-policy scanning.
- Keep the app's internal AI assistant on the builtin provider while delegating chat to an external assistant that calls back through MCP.
Safety and Privacy
Sure MCP is useful because it gives agents structured access to sensitive
finance data. That same access makes it high risk. Scope tokens to trusted
clients, keep the endpoint behind TLS and a private network where possible, and
prefer the Pipelock reverse proxy for inbound MCP traffic. The compose comments
explicitly warn that /mcp remains reachable on the web port when that port is
published, so network design still matters.
Use human confirmation for any tool that writes state, especially create_goal
and import_bank_statement. Treat document search and bank-statement extraction
as sensitive data flows because uploaded files and extracted excerpts may reach
model providers, vector stores, external agents, logs, traces, and MCP clients.
Duplicate Check
Checked current content/mcp/, content/tools/, content/agents/,
content/skills/, README entries, open pull requests, and repository-wide
content for Sure Finance, Sure MCP, we-promise/sure, Maybe Finance MCP,
OpenClaw Sure, personal finance MCP, and finance AI MCP. No dedicated Sure MCP
Server entry, exact source URL duplicate, target file, or open duplicate PR was
found.
Disclosure
Editorial listing. No paid placement or affiliate link is used. Sure is AGPL-3.0 open-source software. The hosted Sure site, OpenClaw, Pipelock, OpenAI-compatible providers, Anthropic support, vector stores, Langfuse, banking providers, app hosting, and mobile/debug builds may have separate terms, costs, security boundaries, and privacy controls.
Source citations
Add this badge to your README
How it compares
Sure MCP Server side by side with 3 alternatives on trust, install, platform support, and disclosed safety notes — all from reviewed registry metadata.
| Field | Sure MCP Server Self-hosted MCP endpoint built into Sure Finance for external AI assistants that need token-authenticated access to personal finance data, accounts, transactions, holdings, budgets, balance sheets, income statements, uploaded family files, bank-statement imports, and goal creation. Open dossier | Daloopa MCP Server for Claude Access high-quality fundamental financial data from SEC filings and investor presentations Open dossier | Drillr - The Financial MCP for AI Agents MCP Server Streamable HTTP MCP server for financial research workflows, including standardized financial data, SEC filing search, company discovery, market signals, ticker resolution, and alternative-data table exploration. Open dossier | Financial Datasets MCP Server Python MCP server that lets Claude query Financial Datasets for stock statements, prices, company news, crypto prices, and SEC filings. Open dossier |
|---|---|---|---|---|
| Trust | ||||
| Install risk | Review first | Low risk | Review first | Review first |
| Notes | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ | Safety ✓ Privacy ✓ |
| Category | mcp | mcp | mcp | mcp |
| Source | source-backed | first-party | source-backed | source-backed |
| Author | Sure | Daloopa | Drillr | Financial Datasets |
| Added | 2026-06-18 | 2025-09-18 | 2026-06-05 | 2026-06-05 |
| Platforms | Claude CodeCursorClaude Desktop | Claude CodeClaude Desktop | Claude CodeClaude Desktop | Claude CodeClaude Desktop |
| Source repo | — | — | — | — |
| Safety notes | ✓Sure MCP exposes personal finance tools over a token-authenticated endpoint. Treat the bearer token as access to the configured user's family finance data. Current code accepts Doorkeeper bearer tokens with `read_write` scope and also supports `MCP_API_TOKEN` plus `MCP_USER_EMAIL` as an env-token fallback. Most tools read financial data, but `create_goal` creates a Sure goal and `import_bank_statement` can create a transaction import from an uploaded bank-statement PDF. The AI compose file recommends routing external AI traffic through Pipelock's MCP reverse proxy on port 8889, but it also notes that `/mcp` remains reachable on the web port when published. The source version checked during review was `0.7.2-alpha.7`, while the latest GitHub release was `v0.7.1-hotfix.1`; verify deployment maturity before relying on it for production financial workflows. Do not connect untrusted MCP clients, broad autonomous agents, or public endpoints to a Sure instance containing real account, transaction, investment, tax, or document data. | ✓Treat financial data as informational research and verify figures before using them for investment, accounting, or reporting decisions. | ✓Drillr's MCP endpoint is a remote Streamable HTTP server at `https://gateway.drillr.ai/mcp/data` and requires a bearer API key. Tool calls can consume Drillr credits or billable API usage. Check plan limits and remaining credit balance before high-volume research loops. run_sql is documented as read-only SELECT access, but broad queries across 90-plus tables can still be expensive, slow, or noisy. Treat outputs as research inputs, not investment, legal, accounting, tax, or trading advice. Drillr documents that it does not place trades, manage brokerage positions, expose options chains, or provide its own price forecasts. Review cited SEC filing paragraphs, table schemas, ticker resolution, and source coverage before relying on results in reports or decisions. | ✓This server retrieves financial market data; do not treat tool output or model summaries as investment, trading, tax, or legal advice. Current prices, crypto prices, filings, and news can be delayed, incomplete, rate limited, or unavailable depending on API plan and market conditions. Review generated analysis before using it in portfolio decisions, trading workflows, client reports, or compliance-sensitive research. The server sends ticker symbols, date ranges, filing filters, and research queries to the Financial Datasets API. |
| Privacy notes | ✓MCP tools can expose account names, balances, linked-provider status, transaction history, merchants, categories, tags, holdings, securities, budgets, income, expenses, net worth, goals, and family file search results. The bank-statement import tool can send uploaded PDF content to the configured LLM provider for extraction and then create an import for review. The family-file search tool can expose excerpts from uploaded documents such as tax returns, bank statements, contracts, insurance policies, investment reports, CSVs, spreadsheets, and other stored files. OpenAI-compatible providers, Anthropic support, external assistants, OpenClaw, Pipelock, vector stores, Langfuse tracing, proxy logs, Rails logs, OAuth tokens, and bearer headers may all become part of the data-flow boundary depending on configuration. Keep `MCP_API_TOKEN`, OAuth access tokens, `EXTERNAL_ASSISTANT_TOKEN`, provider keys, document excerpts, transaction exports, and Pipelock logs out of public prompts, issues, screenshots, and commits. | ✓Company queries, filing references, financial models, and research topics may be sent through the MCP client and model. | ✓Prompts, ticker lists, research themes, SQL queries, and retrieved financial context are sent to Drillr's hosted gateway and the connected MCP client. Drillr API keys are sensitive credentials. Keep them out of prompts, shared configs, issue comments, logs, screenshots, and repository files. Research prompts may reveal portfolio interests, watchlists, investment theses, client names, diligence topics, or confidential strategy. SEC filing excerpts, analyst consensus, alt-data, market signals, and generated summaries may be retained by MCP clients, AI providers, logs, or downstream tools. Use synthetic prompts or public tickers for demos, screenshots, and examples when client, portfolio, or internal research strategy is sensitive. | ✓API keys must be stored in environment variables or local secret managers and never committed to configuration files. Queried tickers, crypto symbols, time ranges, filings, and news requests can reveal research interests, watchlists, client assignments, or trading hypotheses. Tool outputs may include financial statements, prices, filings, and news that become part of the model context and logs. Avoid sending private portfolio holdings, customer identifiers, or material non-public information through prompts or tool arguments. |
| Prerequisites |
|
|
|
|
| Install | | | | |
| Config | | | | |
| Citations | ||||
| Claim | Unclaimed | Unclaimed | Unclaimed | Unclaimed |
Signals
Loading live community signals…
A short, calm digest of reviewed Claude resources. Unsubscribe any time.