umtri-mcp

MCP serverAI & models

umtri gives your AI a persistent map of your project's structure, dependencies, and bugs. Once added, your AI can see how your code fits together and keep that understanding over time, so it does not start from scratch each session.

Use umtri-mcp in Claude, ChatGPT or Ahel Desktop

Free. Sign in, add umtri-mcp and connect your AI. About a minute.

Also: Claude Code · Cursor · Codex

Then ask your AI: use umtri-mcp

Details

Available today. Use it from your connected AI after setup.

Add umtri, then ask your AI to map your project so it can start building its picture of your code.

umtri-mcpStart free

What your AI can do with it

  • Map your project's structure
  • Track your project's dependencies
  • Keep a record of known bugs
  • Remember project details between sessions

Install umtri-mcp

The server’s own address, for the clients that take one directly. Or connect Ahel once and every client you use reads it from one address, with the account kept on Ahel rather than in each client’s config.

  • Claude Code

    claude mcp add --transport http --scope user umtri-mcp 'https://mcp.umtri.io'

    Run it once in your project, then open /mcp to approve any sign-in the server asks for.

  • Claude Desktop

    https://mcp.umtri.io

    Add a custom connector in Settings, paste this address, and approve the sign-in.

  • Cursor

    cursor://anysphere.cursor-deeplink/mcp/install?name=umtri-mcp&config=eyJ1cmwiOiJodHRwczovL21jcC51bXRyaS5pbyJ9

    Open the link and Cursor adds the server at that address.

  • ChatGPT

    https://mcp.umtri.io

    In Settings, enable Developer mode, create an MCP app, and paste this address. Your plan and workspace must allow custom apps.

  • Codex

    codex mcp add umtri-mcp --url 'https://mcp.umtri.io'

    Run it once, then sign in with codex mcp login umtri-mcp if the server asks for an account.

From the project's README

As published by bepuljang/umtri-mcp in README.md.

MCP server for Umtri — give your AI coding agent a persistent map of your project's structure, dependencies, and bugs.

Your agent re-derives your codebase from scratch every session. Umtri stores the structure once and hands it back through MCP: a tree of what the system is made of, the calls and dependencies between its parts, and the bugs eroding it. Ask "what breaks if I change this?" and get an answer traced through recorded connections rather than guessed from a grep.

Quick start

The fastest path is the hosted endpoint at https://mcp.umtri.io — nothing to install. Generate a token at app.umtri.io → Settings → API Tokens, then:

claude mcp add --transport http umtri https://mcp.umtri.io \
  --header "Authorization: Bearer umtri_pat_xxxxxxxxxxxxxxxx"

Any MCP client that takes a config object uses the same shape:

{
  "mcpServers": {
    "umtri": {
      "type": "http",
      "url": "https://mcp.umtri.io",
      "headers": { "Authorization": "Bearer umtri_pat_xxxxxxxxxxxxxxxx" }
    }
  }
}

Run it yourself (stdio)

Use this package when you want to pin a version or point at a self-hosted API:

claude mcp add umtri \
  -e UMTRI_API_TOKEN=umtri_pat_xxxxxxxxxxxxxxxx \
  --transport stdio \
  -- npx -y umtri-mcp
VariableRequiredDefault
UMTRI_API_TOKENyes—
UMTRI_API_BASEnohttps://api.umtri.io

The token is not validated at setup time, so a bad token connects but every call fails. If tools appear and then error, re-check the token and its scope — read tokens cannot call any write tool.

Tools

Read — get_graph, get_bug, get_impact, list_projects, list_bugs, list_seasons, list_events

Write — create_project, create_node, update_node, delete_node (a subtree only with cascade: true), create_edge, delete_edge, create_api, update_api, delete_api, create_bug, update_bug, delete_bug, create_season (acts only after the user confirms)

Plan loop — commit_plan, record_commit

Wiki — list_wiki, get_wiki, list_wiki_revisions, write_wiki, delete_wiki. The ground's encyclopedia: pages form a table of contents (parent, position) read from an overview down, each has a kind, and every write comes back with warnings when a page strays from the skeleton.

Feedback — send_feedback files a suggestion or defect about Umtri itself.

get_graph returns a slice, not a dump: scope by subtree (rootId), by layer (maxType), by role, or by season, and control description weight separately. A large tree stays cheap to read.

Resources

Ten read-only documents the agent can pull for domain rules — the plant vocabulary (trunk/limb/twig/leaf/vein), what counts as a node and what does not, how seasons work, how plan nodes are meant to be realized, how a ground behaves while transplanting, keeping the tree in step with commits, and how to write and organize the wiki.

umtri://rules/vocabulary            umtri://rules/plan
umtri://rules/vocabulary-detailed   umtri://rules/commit-sync
umtri://rules/system-structure      umtri://rules/wiki
umtri://rules/seasons               umtri://templates/wiki
umtri://rules/transplant            umtri://about/vision

Links

License

Apache-2.0. See LICENSE.

Signals

Last commit
Oct 2026
Weekly downloads
1k
Weekly_downloads
1k weekly_downloads
Advanced
Delivery
umtri MCP server → your Ahel connector (mcp.ahel.ai) → your AI.
Item type
mcp-server
Key
io-umtri-umtri
Source
github.com/bepuljang/umtri-mcp
Hosted endpoint
https://mcp.umtri.io