About Noeto
Noeto is an MCP server for AI agents working with a Noeto kanban board. It supports board and card search, card updates and comments, and documents and attachments, while preventing deletion of user-created content.
Key features
- 01List boards and inspect columns and cards
- 02Search cards by text and attributes
- 03Create, update, and move cards
- 04Comment on cards
- 05Read and attach card documents
- 06Read or download attachments
Use cases
- 01Find cards by assignee, label, or due date
- 02Move a card and update its details
- 03Add a comment or document to a card
- 04Retrieve an attachment from a card
MCP support
The local MCP server exposes read and write tools for one Noeto team, selected by its access token. It can run through Docker or a native binary and connects to Claude Code, Claude Desktop, and other MCP hosts.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
docker run -i --rm -e NOETO_TOKEN -e NOETO_API_URL ghcr.io/noeto-tasks/noeto-mcp:v0.6.5NOETO_TOKENSecretPersonal access token, from Settings → Access tokens in Noeto. It decides which team the server works.NOETO_API_URLThe Noeto API root.Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
ghcr.io/noeto-tasks/noeto-mcp:v0.6.5stdioTools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project READMEnoeto-mcp
An MCP server that lets an AI agent work a noeto kanban board: read it, create and update cards, move them, comment, and keep a written document on a card.

Quick start (Claude Code)
-
In noeto, Settings → Access tokens → Create token, and copy the secret.
-
Install the plugin. It bundles the server and the
/cardworkflow:/plugin marketplace add noeto-tasks/noeto-mcp /plugin install noeto@noeto-mcp -
Export the token and restart Claude Code:
export NOETO_TOKEN=noeto_pat_… export NOETO_API_URL=https://api.noeto.online/api/v1
Then ask: "What's on my board?" More in the plugin README.
Other ways to run it
Docker, for any MCP host:
claude mcp add noeto -s user \
-e NOETO_TOKEN=noeto_pat_… \
-e NOETO_API_URL=https://api.noeto.online/api/v1 \
-- docker run -i --rm -e NOETO_TOKEN -e NOETO_API_URL ghcr.io/noeto-tasks/noeto-mcp:v0.6.5
Or by hand in ~/.claude.json, .mcp.json or Claude Desktop's config:
{
"mcpServers": {
"noeto": {
"command": "docker",
"args": ["run", "-i", "--rm", "-e", "NOETO_TOKEN", "-e", "NOETO_API_URL",
"ghcr.io/noeto-tasks/noeto-mcp:v0.6.5"],
"env": {
"NOETO_TOKEN": "noeto_pat_…",
"NOETO_API_URL": "https://api.noeto.online/api/v1"
}
}
}
}
Native binary, with Homebrew:
brew install noeto-tasks/tap/noeto-mcp
claude mcp add noeto -s user \
-e NOETO_TOKEN=noeto_pat_… \
-e NOETO_API_URL=https://api.noeto.online/api/v1 \
-- "$(brew --prefix)/bin/noeto-mcp"
Or take an archive from releases
(macOS, Linux, Windows). The binary is unsigned, so on macOS run
xattr -d com.apple.quarantine noeto-mcp first. Point the config at it by
absolute path.
Good to know
- One server serves one team — the token decides which. For two teams, add two
entries with their own tokens (
noeto-work,noeto-personal). - Against a local noeto from Docker, use
http://host.docker.internal:8081/api/v1(macOS, Windows) or--network host(Linux). - It runs as a local process, so it works in Claude Code and Claude Desktop, not in claude.ai in the browser or the mobile apps.
Tools
| tool | what it does | |
|---|---|---|
list_boards | the team's boards, with card counts | reads |
get_board | one board: columns in order and the cards in them | reads |
find_cards | search every board by text, assignee, label, priority, column, due date | reads |
get_card | one card in full, with comments and the files on it | reads |
list_members | who is on the team | reads |
whoami | which member the token belongs to | reads |
create_card | add a card to a column | writes |
update_card | title, description, assignee, priority, due date, labels | writes |
move_card | to another column, or reorder within one | writes |
comment_on_card | post a comment | writes |
read_document | the Markdown of a document on a card | reads |
attach_document | write one, replacing the previous copy of that name | replaces |
read_attachment | a text or image file on a card | reads |
download_attachment | save any file on a card to disk and return its path | saves locally |
- Boards, columns, labels and people can be named:
move_card(card, column: "Done"). The card itself needs its id, so a typo can never edit the wrong card. noneclears a field:update_card(card, due: "none"). An omitted argument leaves it alone.- Documents and text files are capped at 512 KB, images at 1.5 MB. Bigger files
and other types go through
download_attachment. - Downloads land in
NOETO_DOWNLOAD_DIR(default:noeto-attachmentsin the system temp directory). In Docker, mount that directory at the same path on both sides — the plugin's config already does.
It deliberately cannot delete anything a person made, and does not create boards, columns or labels. The reasoning behind these and the other choices is in docs/internals.md.
Development
make test # hermetic, against a fake API
make smoke # contract check against a running noeto (needs NOETO_TOKEN)
make lint
make docker # build the image for this machine
make release-dry # build release artifacts into dist/, publish nothing
Run make smoke after any API change.
Releasing: Actions → Release → Run workflow, with the version (0.6.3).
It tests, pins the version, pushes the image, tags, publishes the GitHub release
and Homebrew cask, and lists the version in the MCP Registry. It needs a
HOMEBREW_TAP_TOKEN secret that can write to noeto-tasks/homebrew-tap.
make release-plugin RELEASE=x.y.z does the same from a laptop, except the
registry: run the Publish to MCP Registry workflow afterwards.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.