Monarch

mikelane5stars

An agentic budgeting companion — a local-first MCP server that turns Claude into a financial advisor over your Monarch Money data (read + categorize only).

Finance

Install Run on PlugMCP

GitHub

Run on PlugMCP

1# Monarch

README.mdMCP server

Transport: stdio
Status: Local

Power your agent

Add this server

Install from the publisher. Not a PlugMCP endpoint.

Run on PlugMCP

Host a copy with our keys and logs.

No packaged release in the registry yet. Follow the repository README.

RepositoryPublisher installStep-by-step: add this in Claude

Remote URL or a command on your machine? How the two installs differ

Run on PlugMCP

Host a copy with our keys and logs.

A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.

Launch

Server facts

Local

Transport: stdio
License: Apache-2.0
Last checked: 2026-10-28

Stars: 5
Tools: —
Uptime: —

Tools

No tool list yet. Remote servers get one after the first health check.

FAQ

Questions

How do I install Monarch?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does Monarch need an API key?

No required environment variable is listed on this record.

What transport does Monarch use?

Monarch runs locally over stdio.

When was Monarch last checked?

Last checked 2026-10-28. An agentic budgeting companion — a local-first MCP server that turns Claude into a financial advisor over your Monarch Money data (read + categorize only).

From the publisher

Documentation

monarch-mcp

An agentic budgeting companion. A local-first Model Context Protocol server that turns Claude (or any MCP client) into a financial advisor over your Monarch Money data — it knows where your household stands, keeps the books honest, measures progress against your goals, and looks ahead.

[!IMPORTANT] Unofficial. Monarch Money has no public API. This server talks to Monarch's private GraphQL API and is not affiliated with, endorsed by, or supported by Monarch Money. Use at your own risk. See DISCLAIMER.md.

Why this exists

Monarch has no official API, so every Claude↔Monarch integration is a community-built wrapper. Rather than hand a third-party server full access to a live banking session, this is a purpose-built server you can read end to end, with one defining property:

It cannot move money — by construction. There is no transfer, payment, withdrawal, create, or delete code anywhere in it. The only write path can change a transaction's category, tags, or notes — nothing else. Capability denial is enforced by the code's absence of those tools, not by configuration you can misset.

It's read-and-categorize only, your session token never leaves your machine, and the whole thing is validated against real Monarch by a proper test pyramid.

The tools

Compound, task-oriented tools — each does a job, combining several API calls + computation into one decision-ready result (not 1:1 API wrappers you could replace with curl):

Tool	What it answers
`financial_overview`	"Where do we stand?" — net worth, this-month cash flow, month-over-month change
`spending_report`	Spending by category vs budget, over-budget flags, likely duplicates, vs prior period
`progress_vs_goals`	Actuals vs your stored goals — on track / drifting / off
`cashflow_forecast`	Upcoming bills + income → projected month-end position + shortfall warnings
`net_worth_trend`	Net worth over time, deltas by account type, biggest movers
`recurring_scan`	New / changed / "creeping" subscriptions and upcoming renewals
`account_inventory`	All accounts bucketed by retirement-planning role — tax-advantaged, taxable brokerage, cash, other assets, liabilities — with a net-worth rollup
`triage_uncategorized` + `apply_changeset`	Propose categories from your own history, then commit only what you approve

Quick start

1. Install

# One-line shell installer (macOS / Linux)
curl --proto '=https' --tlsv1.2 -LsSf \
  https://github.com/mikelane/monarch-mcp/releases/latest/download/monarch-mcp-installer.sh | sh

# PowerShell (Windows)
irm https://github.com/mikelane/monarch-mcp/releases/latest/download/monarch-mcp-installer.ps1 | iex

# Cargo, prebuilt binary — no compile (fetches the release artifacts)
cargo binstall monarch-mcp

# Cargo (from crates.io) — compiles from source
cargo install monarch-mcp

# Or build from source
git clone https://github.com/mikelane/monarch-mcp
cd monarch-mcp
cargo install --path .

2. Authenticate (one time)

monarch-mcp login

Prompts for your Monarch email, password, and MFA code. The session token is written to ~/.config/monarch-mcp/session.json (mode 0600) and reused for months — your password and MFA secret are never stored.

Sign in with Apple/Google? Monarch's API needs email+password. Add a password in Monarch → Settings → Security; your SSO login keeps working alongside it.

3. Register with your MCP client

Claude Code (works today, local stdio):

claude mcp add monarch-mcp -- monarch-mcp

…or copy .mcp.json.example to .mcp.json in your project.

Streamable-HTTP (local, loopback-only):

monarch-mcp --http

Starts the same tool set over streamable-HTTP MCP instead of stdio, serving /mcp on 127.0.0.1:8770 by default. Override the bind address with MONARCH_HTTP_ADDR (e.g. MONARCH_HTTP_ADDR=127.0.0.1:9000 monarch-mcp --http). A /healthz endpoint returns ok for basic liveness checks.

[!IMPORTANT] This mode refuses to bind any non-loopback address — a hard error, never a silent fallback. It also rejects any request that carries a non-localhost Origin header (an absent Origin, the normal case for non-browser MCP clients, is allowed), which stops DNS-rebinding (a malicious webpage rebinding an attacker hostname to 127.0.0.1 so your browser POSTs to localhost:8770); loopback binding alone doesn't catch that. See ADR 0018. There's still no auth and no tunnel here — this is Phase 1 of Claude Cowork support (issue #88). Exposing the server beyond your machine is Phase 2, and it doesn't exist yet.

[!NOTE] Claude Cowork runs MCP servers in an isolated VM and currently can't reach a local stdio server (known issue). --http above is the first step toward supporting Cowork, but it's loopback-only — Cowork's VM still can't reach it without the Phase 2 tunnel work. Use it from Claude Code on your machine for now. See docs for the trade-offs.

4. Use it

"Give me a financial overview." · "How's our spending vs budget this month?" · "Any creeping subscriptions?" · "Are we on track for our goals?"

Goals (optional)

progress_vs_goals measures against a TOML file you point to with MONARCH_GOALS_FILE. See goals.example.toml:

[savings_rate]
target_percent = 20.0

[emergency_fund]
target_months = 6.0

Security model

No money movement exists in the binary. The mutating path is an allowlist of category / tags / notes; any other field in a change request is rejected and reported.
Your token stays local. Auth happens on your machine; the session lives in ~/.config/monarch-mcp/session.json (0600). Credentials are never logged.
Human-in-the-loop writes. triage_uncategorized only proposes; nothing is written until you approve a changeset and apply_changeset commits exactly that.

Reporting a vulnerability: see SECURITY.md.

How it's built

Production Rust (rmcp + reqwest + tokio), built TDD/BDD-first, with a deliberate Google test-size pyramid — because testing only against mocks ships false confidence (it did, twice — see the ADRs):

Tier	What	Count
Small	Hermetic, in-process unit tests (the aggregation/classification math)	~270
Medium	Behave BDD against a mock Monarch GraphQL server over loopback	~57
Large	Gated (`MONARCH_LIVE=1`) integration tests against real Monarch	8

cargo test                                   # small
cd bdd && uv run behave                       # medium (needs uv + the built binary)
MONARCH_LIVE=1 cargo test --test live_integration   # large (needs a real session)

The real GraphQL schema and the design decisions are documented in docs/decisions/ (ADRs 0001–0003).

Project layout

src/            Rust MCP server — client, tools, goals, server, error
bdd/            Python + behave acceptance suite + mock Monarch GraphQL server
tests/          Large/live integration tests (gated)
docs/decisions/ ADRs (auth flow, real schema, tier-2 schema)
docs/specs/     Design spec

Contributing

Contributions welcome — see CONTRIBUTING.md for the dev setup (mise / lefthook / clippy / rustfmt), the test-pyramid expectations, and the TDD/BDD workflow. Also read CODE_OF_CONDUCT.md.

For the release process (secrets, merge→publish flow), see docs/RELEASING.md.

License

Dual-licensed under either of MIT or Apache-2.0 at your option. Unless you explicitly state otherwise, any contribution you submit shall be dual-licensed as above, without additional terms.

Monarch

mikelane5stars

An agentic budgeting companion — a local-first MCP server that turns Claude into a financial advisor over your Monarch Money data (read + categorize only).

Finance

Install Run on PlugMCP

GitHub

Run on PlugMCP

1# Monarch

README.mdMCP server

Transport: stdio
Status: Local

Power your agent

Add this server

Install from the publisher. Not a PlugMCP endpoint.

Run on PlugMCP

Host a copy with our keys and logs.

No packaged release in the registry yet. Follow the repository README.

RepositoryPublisher installStep-by-step: add this in Claude

Remote URL or a command on your machine? How the two installs differ

Run on PlugMCP

Host a copy with our keys and logs.

A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.

Launch

Server facts

Local

Transport: stdio
License: Apache-2.0
Last checked: 2026-10-28

Stars: 5
Tools: —
Uptime: —

Tools

No tool list yet. Remote servers get one after the first health check.

FAQ

Questions

How do I install Monarch?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does Monarch need an API key?

No required environment variable is listed on this record.

What transport does Monarch use?

Monarch runs locally over stdio.

When was Monarch last checked?

Last checked 2026-10-28. An agentic budgeting companion — a local-first MCP server that turns Claude into a financial advisor over your Monarch Money data (read + categorize only).

From the publisher

Documentation

monarch-mcp

[!IMPORTANT] Unofficial. Monarch Money has no public API. This server talks to Monarch's private GraphQL API and is not affiliated with, endorsed by, or supported by Monarch Money. Use at your own risk. See DISCLAIMER.md.

Why this exists

It cannot move money — by construction. There is no transfer, payment, withdrawal, create, or delete code anywhere in it. The only write path can change a transaction's category, tags, or notes — nothing else. Capability denial is enforced by the code's absence of those tools, not by configuration you can misset.

It's read-and-categorize only, your session token never leaves your machine, and the whole thing is validated against real Monarch by a proper test pyramid.

The tools

Compound, task-oriented tools — each does a job, combining several API calls + computation into one decision-ready result (not 1:1 API wrappers you could replace with curl):

Tool	What it answers
`financial_overview`	"Where do we stand?" — net worth, this-month cash flow, month-over-month change
`spending_report`	Spending by category vs budget, over-budget flags, likely duplicates, vs prior period
`progress_vs_goals`	Actuals vs your stored goals — on track / drifting / off
`cashflow_forecast`	Upcoming bills + income → projected month-end position + shortfall warnings
`net_worth_trend`	Net worth over time, deltas by account type, biggest movers
`recurring_scan`	New / changed / "creeping" subscriptions and upcoming renewals
`account_inventory`	All accounts bucketed by retirement-planning role — tax-advantaged, taxable brokerage, cash, other assets, liabilities — with a net-worth rollup
`triage_uncategorized` + `apply_changeset`	Propose categories from your own history, then commit only what you approve

Quick start

1. Install

# One-line shell installer (macOS / Linux)
curl --proto '=https' --tlsv1.2 -LsSf \
  https://github.com/mikelane/monarch-mcp/releases/latest/download/monarch-mcp-installer.sh | sh

# PowerShell (Windows)
irm https://github.com/mikelane/monarch-mcp/releases/latest/download/monarch-mcp-installer.ps1 | iex

# Cargo, prebuilt binary — no compile (fetches the release artifacts)
cargo binstall monarch-mcp

# Cargo (from crates.io) — compiles from source
cargo install monarch-mcp

# Or build from source
git clone https://github.com/mikelane/monarch-mcp
cd monarch-mcp
cargo install --path .

2. Authenticate (one time)

monarch-mcp login

Sign in with Apple/Google? Monarch's API needs email+password. Add a password in Monarch → Settings → Security; your SSO login keeps working alongside it.

3. Register with your MCP client

Claude Code (works today, local stdio):

claude mcp add monarch-mcp -- monarch-mcp

…or copy .mcp.json.example to .mcp.json in your project.

Streamable-HTTP (local, loopback-only):

monarch-mcp --http

[!IMPORTANT] This mode refuses to bind any non-loopback address — a hard error, never a silent fallback. It also rejects any request that carries a non-localhost Origin header (an absent Origin, the normal case for non-browser MCP clients, is allowed), which stops DNS-rebinding (a malicious webpage rebinding an attacker hostname to 127.0.0.1 so your browser POSTs to localhost:8770); loopback binding alone doesn't catch that. See ADR 0018. There's still no auth and no tunnel here — this is Phase 1 of Claude Cowork support (issue #88). Exposing the server beyond your machine is Phase 2, and it doesn't exist yet.

[!NOTE] Claude Cowork runs MCP servers in an isolated VM and currently can't reach a local stdio server (known issue). --http above is the first step toward supporting Cowork, but it's loopback-only — Cowork's VM still can't reach it without the Phase 2 tunnel work. Use it from Claude Code on your machine for now. See docs for the trade-offs.

4. Use it

"Give me a financial overview." · "How's our spending vs budget this month?" · "Any creeping subscriptions?" · "Are we on track for our goals?"

Goals (optional)

progress_vs_goals measures against a TOML file you point to with MONARCH_GOALS_FILE. See goals.example.toml:

[savings_rate]
target_percent = 20.0

[emergency_fund]
target_months = 6.0

Security model

No money movement exists in the binary. The mutating path is an allowlist of category / tags / notes; any other field in a change request is rejected and reported.
Your token stays local. Auth happens on your machine; the session lives in ~/.config/monarch-mcp/session.json (0600). Credentials are never logged.
Human-in-the-loop writes. triage_uncategorized only proposes; nothing is written until you approve a changeset and apply_changeset commits exactly that.

Reporting a vulnerability: see SECURITY.md.

How it's built

Tier	What	Count
Small	Hermetic, in-process unit tests (the aggregation/classification math)	~270
Medium	Behave BDD against a mock Monarch GraphQL server over loopback	~57
Large	Gated (`MONARCH_LIVE=1`) integration tests against real Monarch	8

cargo test                                   # small
cd bdd && uv run behave                       # medium (needs uv + the built binary)
MONARCH_LIVE=1 cargo test --test live_integration   # large (needs a real session)

The real GraphQL schema and the design decisions are documented in docs/decisions/ (ADRs 0001–0003).

Project layout

src/            Rust MCP server — client, tools, goals, server, error
bdd/            Python + behave acceptance suite + mock Monarch GraphQL server
tests/          Large/live integration tests (gated)
docs/decisions/ ADRs (auth flow, real schema, tier-2 schema)
docs/specs/     Design spec

Contributing

Contributions welcome — see CONTRIBUTING.md for the dev setup (mise / lefthook / clippy / rustfmt), the test-pyramid expectations, and the TDD/BDD workflow. Also read CODE_OF_CONDUCT.md.

For the release process (secrets, merge→publish flow), see docs/RELEASING.md.

License

Dual-licensed under either of MIT or Apache-2.0 at your option. Unless you explicitly state otherwise, any contribution you submit shall be dual-licensed as above, without additional terms.