amazing-marvin-mcp
Enables AI assistants to query and modify Amazing Marvin tasks, projects, labels, and subtasks, using a local mirror for fast reads and the official REST API for writes.
README
amazing-marvin-mcp
Overview
An unofficial MCP server for Amazing Marvin, so an assistant such as Claude can answer "what's in my inbox?", "what's due this week?", "I'm at the car, what can I do?", and act on it: rename, relabel, reschedule, move, split into projects, save research into a note.
This project is not affiliated with, endorsed by, or supported by Amazing Marvin; it uses their public API.
How it works
Claude ──MCP (stdio)──▶ amazing-marvin-mcp
├─ local mirror ◀── CouchDB `_changes` (your sync database)
│ reads, search and filters run here: no rate-limit cost
└─ writes ──▶ Marvin REST API (addTask, doc/update, markDone)
throttled to Marvin's limits (1 request / 3 s)
Marvin's REST API has no "list" or "search" endpoint and a budget of 1440
requests per day, so reads come from a local copy of your database kept fresh
with an incremental _changes poll (at most once a minute). Writes go through
the REST API, which handles Marvin's conflict-resolution bookkeeping
(fieldUpdates) server-side.
What the assistant knows
- Marvin semantics, in the UI's own words. Inbox = not filed anywhere; categories and projects nest arbitrarily (the Master List). The four dates are kept apart: Do date (the day you plan to do it, "scheduled"), Due date (hard external deadline), End date (self-imposed target; Planning Ahead's planned week/month is the softer form), Start date (hidden on the Backburner until then). Plus Duration estimates (minutes), Importance (P1–P3 stars), Eat-the-Frog, Backburner, labels and label groups.
- Your structure (
get_structure): the category/project tree with ids and open-task counts, labels by group, and which Marvin strategies you have enabled (by their names in the Strategies screen). - Your workflow (
workflow.mdor aworkflow/directory of sections): how you use Marvin: what labels mean, how you plan, your inbox-triage checklist, what "short win" means. Marvin is flexible, so this is what lets the assistant act the way you would. It is injected into the server instructions and exposed asmarvin://workflow(andmarvin://workflow/<section>).
Tools
| Tool | What it does |
|---|---|
get_structure |
Category/project tree with ids, labels by group, strategies in use |
list_inbox |
Open inbox tasks, oldest first |
list_today [day] |
Do date that day, do date earlier but not done, due by then |
list_due [by] |
Open tasks with a due date by a date (week, month, YYYY-MM-DD) |
search_tasks … |
Any mix of text, parent, labels, do-date window, due/end by, estimate, flags |
get_task id |
Full detail of a task (note, subtasks, dates) or a project (its open children) |
list_children parent |
Direct tasks and sub-projects of a project/category |
create_task / create_project |
Names for parent/labels are resolved for you |
create_category / create_label |
New folder in the Master List; new label (in an existing or new group) |
update_subtasks id … |
Add, complete, reopen, rename, remove subtasks |
update_task id … |
Rename, move, relabel, do/due/end/start date, estimate, note, importance, clear… |
mark_done id |
Complete a task/project |
sync_marvin |
Force a mirror refresh |
Prompts: triage_inbox, daily_review |
Built from your workflow file |
There is deliberately no delete tool: Marvin's trash is client-side, so API deletes are unrecoverable.
Install and set up (using it)
-
Install (needs uv):
uv tool install git+https://github.com/tsvikas/amazing-marvin-mcp.git -
Credentials. In Marvin, enable the API strategy (Strategies → API → settings). Put its values in
~/.config/amazing-marvin-mcp/.env(Linux;amazing-marvin-mcp checkprints the exact path on your OS):MARVIN_API_TOKEN=... # create / mark done MARVIN_FULL_ACCESS_TOKEN=... # edit existing items; omit for no edits MARVIN_SYNC_SERVER=... # CouchDB: the read mirror MARVIN_SYNC_DATABASE=... MARVIN_SYNC_USER=... MARVIN_SYNC_PASSWORD=...Environment variables and a
.envin the working directory also work (and override the per-user file). While trying things out, use a second, throwaway Marvin account: the full-access token can damage data. -
Check and first sync:
amazing-marvin-mcp check # verifies tokens, pulls the database, prints counts -
Describe your workflow:
amazing-marvin-mcp init-workflow # one workflow.md to edit, or amazing-marvin-mcp init-workflow --split # a workflow/ directory: planning, labels, triage, daily…Everything in it goes into the model's instructions (and each section is also a
marvin://workflow/<section>resource), so keep it short and concrete. -
Register with your MCP client.
Claude Code:
claude mcp add marvin -- amazing-marvin-mcp serveClaude Desktop (
claude_desktop_config.json):{ "mcpServers": { "marvin": { "command": "amazing-marvin-mcp", "args": ["serve"] } } }Both read the per-user
.env, so no secrets go in the client config. Then ask: "what's in my Marvin inbox?"
Other settings (shown by check): MARVIN_WORKFLOW_FILE (file or directory), MARVIN_CACHE_DIR,
MARVIN_MIN_REQUEST_INTERVAL (seconds between REST calls, default 3),
MARVIN_MIRROR_MAX_AGE (seconds before a read re-polls, default 60).
The mirror file holds all your tasks; it is written with owner-only permissions under the cache directory.
Developing
git clone https://github.com/tsvikas/amazing-marvin-mcp && cd amazing-marvin-mcp
uv sync && just prepare # deps + pre-commit hooks
just test && just lint
Run the server from the checkout (a .env in the repo root is git-ignored and
is picked up when the client starts from this directory; otherwise use
-e VAR=value or the per-user file):
claude mcp add marvin-dev -- uv run --directory "$PWD" amazing-marvin-mcp serve
tests/test_live.py runs every tool end to end and replays a recorded
cassette by default (tests/cassettes/), which doubles as a fixture of real
Marvin documents. To re-record against a throwaway account (it creates
and then deletes a few [live] items):
MARVIN_API_TOKEN=... MARVIN_FULL_ACCESS_TOKEN=... MARVIN_SYNC_SERVER=... \
MARVIN_SYNC_DATABASE=... MARVIN_SYNC_USER=... MARVIN_SYNC_PASSWORD=... \
uv run pytest tests/test_live.py --record-mode=rewrite
Design rules for contributions are in CLAUDE.md.
Contributing
Interested in contributing? See CONTRIBUTING.md for development setup and guidelines.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.