Skip to main content
Glama
Purser MCP Server

Purser MCP Server

by mlamoure

Purser

The privately hosted finance back-end, designed for use by your personal AI Agent.

Purser syncs bank accounts, categorizes transactions, tracks bills, reconciles statements, forecasts cash flow and values investments. It is mostly headless: you work through your own AI agent over MCP or REST. A web UI provides the administrative controls for setup, configuration, access and record inspection.

Purser is not an AI. It is an AI enablement engine: it stores financial data, performs calculations and provides tools for your agent to use. Purser requires no AI subscription. Optional integrations can call a model you configure for categorization and review; they are off by default.

You choose what to share with your agent. Purser provides access scopes, disclosure settings and an audit log; it does not prescribe how much financial information you should share.

Documentation: https://mlamoure.github.io/purser/

Hosting and access

  • Privately hosted. Purser runs on your own hardware with Docker. Bank data is stored on your server, with sensitive fields encrypted at rest. There is no Purser cloud account.

  • Scoped access. Give an agent read access or read and write access. Token creation and configuration require a separate admin scope.

  • Disclosure controls. MCP settings can replace payee names with stable references and omit individual amounts. These are defaults: an agent can request those fields per call. REST reads have different disclosure behavior; see Privacy and agents.

  • Server-side calculations. Tools return totals, balances, net worth and forecasts without requiring the agent to read every transaction. An agent can also link to records for you to inspect in Purser.

  • Auditing. The Data shared view identifies integrations, returned field inventories, coverage, omissions and outcomes, with filters and exports. Older entries and incomplete evidence are labeled explicitly. Integration responses are recorded before release; a capture error withholds the response, although a requested write may already have completed. Posted transactions are corrected through linked reversal and replacement entries.

Related MCP server: finledger

What you can ask your agent

  • "How much did we spend on groceries last month, compared with the month before?"

  • "What's our net worth, and how has it changed this year?"

  • "Will checking stay above $500 until the end of the month?"

  • "Which bills haven't come in yet?"

  • "Here's my card statement email. Set the payment to the statement balance."

  • "Anything new or unusual since yesterday?"

Features

Accounts and imports

  • Bank sync through SimpleFIN Bridge for checking, savings, credit cards and brokerage accounts.

  • QIF and CSV imports with a preview and duplicate detection.

  • Investments: holdings from your brokerage feed or tracked by hand, priced daily and included in net worth.

  • Tax lots and equity awards for holdings you track by hand: load a broker's lot list and vest schedule in one previewed step. Sales name the lots they draw on. Home includes potential value inside the Net worth and Assets cards when unvested awards exist; users without them see only current owned totals. Unvested awards remain outside owned net worth until vesting.

Ledger and reconciliation

  • Balanced entries. Transaction postings sum to zero per currency. Corrections to posted transactions preserve the original entry and its history.

  • Source records. Original bank payloads are retained for tracing imported transactions.

  • Reconciliation against account statements, including explicit cash-only reconciliation for manually tracked brokerage accounts.

  • Transfers between your own accounts are linked and excluded from spending totals.

Bills and forecasts

  • Recurring bills and income are detected or entered as schedules, then matched to incoming transactions.

  • Occurrence estimates: amount and date uncertainty are tracked separately, with evidence, original estimates and linked actuals. Fixed planning values can remain estimates, including projected slots without transaction IDs, without changing forecast amounts or dates. Confirm one payment without confirming later ones; ambiguous matches wait for review.

  • Late bills and price changes are flagged.

  • Cash-flow forecasts use schedules to project account balances and flag balances below your chosen floor.

Categories, rules and notes

  • Categories come with a default hierarchy you can edit.

  • Rules categorize transactions and rename payees. Draft rules take effect when activated.

  • Review and suggestions inboxes collect new transactions, proposed schedules, transfer links and matches.

  • Financial notes keep context about your finances in one place, such as "the HOA bill is annual," for your agent to use.

The admin interface

Day-to-day work happens through your AI agent. The web UI provides the administrative controls for setup, connections, settings, tokens and auditing. It also lets you inspect accounts and records directly. These screenshots use synthetic demonstration data.

Needs attention with grouped suggestions, account filters and review links

Needs attention: review missing expected payments first, expand suggestions by type, and open the exact account register for posted or pending review.

API and MCP access settings Access: create scoped agent tokens

Privacy with agents settings Disclosure: set defaults for MCP responses

Audit log Audit log: review sharing and changes

Account register with categories and transfer links Record inspection: view transactions, balances and transfer links

Get started

Purser runs with Docker on Linux (amd64 or arm64) and on Macs with Apple Silicon. To have your agent install it, give it this instruction:

Set up Purser for me by following https://mlamoure.github.io/purser/tutorials/ai-setup/

The setup guide asks you to store your encryption key, create the admin account and approve the agent's access token.

For manual installation, follow the Quickstart. Agents that load skills can also use skills/purser/SKILL.md.

Reference

Bugs, ideas and security

Report bugs and ideas in GitHub Issues. Report security problems privately through Report a vulnerability on the Security tab. Pull requests are welcome; the maintainer applies accepted changes upstream.

License

AGPL-3.0-only. Third-party components and their licences are listed in THIRD_PARTY_NOTICES.md.

Release security checks

Release qualification audits the locked Python runtime dependencies and scans native amd64 and arm64 container images. The publication job scans the exact digest again before creating the release tag. High and critical findings block publication unless an unfixed finding has a reviewed, package-version-specific exception in .trivyignore.yaml; exceptions expire and never hide an available fix. Full scan reports remain available as workflow artifacts.

To reproduce locally on Linux, run bash scripts/audit_python.sh and bash scripts/scan_image.sh <image>. Both write to security-reports/.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      D
      maintenance
      Enables AI agents to interact with a double-entry ledger, offering tools for account management, balanced journal entries, balance queries, trial balance, and penny-perfect allocation. Built with safety by construction: no update/delete tools, idempotent posting, and an append-only journal.
      7
      MIT
    • A
      license
      B
      quality
      D
      maintenance
      Enables an agent to query local financial ledgers for expenses, investments, and income, with tools for savings rate, accounts, sync, and reconciliation.
      18
      MIT
    • A
      license
      B
      quality
      B
      maintenance
      Enables AI assistants to query and manage self-hosted accounting data—invoices, balances, and books—through natural language, with read-only tools by default and optional scoped write operations.
      30
      27 PyPI
      1
      AGPL 3.0
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI assistants to read, write, and audit personal finance data over a single SQLite file, covering transactions, receipts (including receipt photos as image blocks), budgets, categories, and line-item price tracking. Every write is an idempotent, soft-deleted, audited upsert so agents can safely transcribe and organize spending.
      27 npm
      MIT