todo-mcp
Enables managing tasks in a structured TODO.md file through MCP tools for listing, adding, updating, setting status, logging notes, searching, and removing tasks while preserving hand-authored formatting.
README
todo-mcp
An MCP server whose store is a TODO.md you can read, edit and diff by hand. Writes are
byte-range splices, so the file stays yours: hand-authored tables, tab indentation and any
prose outside a task are never re-serialised.
Speaks stdio for MCP clients and Streamable HTTP for everything else.
Credit
Based on CalamityAdam/mcp-todo, which supplied
the original scaffold: the createTodoMcpServer factory shape, the Express Streamable HTTP
wrapper and the session handling.
Almost nothing else survives. That version kept todos as numbered records in a JSON blob at
~/.mcp-todos.json with three tools over { id, title, done }. This one replaces the store
with a markdown document, swaps numeric ids for slugs, and grows the tool surface to seven
with statuses, areas, reference breadcrumbs, dated log notes, full-text query and
duplicate detection. The two projects no longer share an implementation.
Upstream ships no LICENSE file; its package.json declares ISC, which is what this repo
carries forward.
Install
Run it straight from GitHub, no clone:
npx github:adrianhardy/todo-mcp
After pushing any changes, use npx --ignore-existing github:adrianhardy/todo-mcp to pick them up.
For regular use, install once and forget about it:
npm i -g github:adrianhardy/todo-mcp
todo-mcp
Either route builds from source on install via the prepare script, so dist/ is never
committed. Node 20 or newer.
Usage
todo-mcp starts the HTTP server by default, because that is the useful thing to do
when a person runs it in a terminal. Set MCP_STDIO=1 to speak stdio instead, which is
what an MCP client spawning it as a subprocess wants.
With an MCP client
{
"mcpServers": {
"todo": {
"command": "npx",
"args": ["-y", "github:adrianhardy/todo-mcp"],
"env": { "MCP_STDIO": "1" }
}
}
}
Installed globally, that becomes "command": "todo-mcp" with the same env block.
The working directory decides which file you get. TODO_FILE resolves against the
process's cwd and defaults to TODO.md, so a client launched in a project edits that
project's todo list. Set TODO_FILE to an absolute path if you want one shared list
regardless of where the server starts.
Over HTTP
PORT=8080 TODO_MCP_TOKEN=$(openssl rand -hex 32) todo-mcp
POST /mcp- JSON-RPC requestsGET /mcp- SSE stream for server notificationsDELETE /mcp- end the session
Setting TODO_MCP_TOKEN requires Authorization: Bearer <token> on all three. Leaving it
unset disables authentication, which is fine on localhost and nowhere else.
Configuration
| variable | default | meaning |
|---|---|---|
TODO_FILE |
TODO.md |
store path, resolved against cwd |
MCP_STDIO |
unset | 1 selects stdio instead of HTTP |
PORT |
3000 |
HTTP port |
TODO_MCP_TOKEN |
unset | bearer token; unset means no auth |
A .env file is read if present. See .env.example.
Storage
TODO.md is the store, not a JSON blob. The file is the record: readable, editable by
hand, and diffable in git.
A task is a ## section. Fields the server owns live in a comment block directly under
the heading; everything below it is prose the human owns.
## Feature Idea version two: the new widget which tracks things
<!-- todo
id: feature-idea-version-two
area: inventory
status: next
refs: [./src/do_stuff.ts, ClassName.Method, OtherClassName]
created: 2026-08-19
updated: 2026-08-22
-->
**Next step:** close the ledger. ClassName.Method uses 0.25 and it needs 17.2%.
**Already known:** ...
### Log
- 2026-08-22 Slab_Wall_1x3 not started; parade places 24 of those to every 6 of the 3x3.
Ids are slugs, not numbers, so they survive reordering and deletion. File order is priority order, which is why there is no priority field.
Writes are byte-range splices: a mutation rewrites only the span it owns. Hand-authored tables, tab indentation and any prose outside a task section are never re-serialised, so they cannot be reflowed or lost. Handlers are serialised through a lock, because two interleaved read-modify-write cycles would splice against offsets that no longer describe the file.
Design
- list is abbreviated
list_todosreturns a one-line index and never task bodies.get_todoreturns one whole section.qorrefis the expected path; listing everything is the exception. - Capture takes one field. Only
titleis required, and new tasks default to statuscaptured. A tool that demands an area and a next step at the moment something is noticed does not get used, and the file only earns its keep if things get written down as they are found. Triage movescapturedtoopen/next/parked/somedaylater.
Status values
| status | meaning |
|---|---|
captured |
raw, untriaged. The default for a new task. Hidden from unfiltered lists |
open |
real work, understood |
next |
up now |
parked |
deliberately deferred; the body says why |
someday |
aspirational |
done |
finished. Stays in the file for the record. Hidden from unfiltered lists |
Tools
list_todos- abbreviated index; filtersarea,status,ref,q,limitget_todo- full markdown of one task, body includedadd_todo- capture a task; onlytitlerequired. Reports possible duplicatesupdate_todo- change any field; only what is passed is rewrittenappend_note- add a dated bullet to a task's logset_status- move a task through triageremove_todo- delete a task and its prose. Preferset_status done
Resources
todos://list- one-line index of open tasks
Architecture
src/todo.ts- the markdown store: parsing, byte-range patching, query, dedupsrc/server.ts- MCP tool surface. Pure factory, no side effects on importsrc/http.ts- Streamable HTTP transport, auth and session mapsrc/cli.ts- thetodo-mcpbinary; picks a transport and starts it
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.
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.
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.
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.