What it enables · 04 / Connect

Feed the systems you already use

Bring Intelligence into Claude, Codex, Cursor, internal agents and enterprise workflows through MCP.

MCPWorkspace
What it is

GitMir publishes what it knows as an MCP server, so the assistant you already use can ask it directly. Seventeen tools cover the same ground as the screens: what this part is, what a change reaches, what the product calls something, what changed between builds, what work is waiting. Claude Code, Codex and Cursor are all good at working in your codebase; GitMir hands them the product understanding they would otherwise rebuild from the repository on every task. The address /mcp?project=<id> starts the assistant in one repository or in the whole project.

What you do
01

Open the repository's Connect MCP tab, or press Connect an AI agent in the project's header.

02

Pick your client: GitMir creates the key and gives one command for Claude Code or Codex, or a one-click button for Cursor and VS Code.

03

Confirm it worked: the client lists eighteen gitmir_ tools.

04

Ask your assistant a question about the product — it calls the tools itself.

What it does not do

Each agent has its own key that reads only the project or repository it was connected to; a key that may not take work can read and ask but is refused when it tries to record a word, take a task or close one. Every call names one repository or one project; there is no call that covers a whole workspace at once. Traffic goes one way — GitMir answers and hands out work, it never reaches into your runtime, your issue tracker or your data. One-click setup is written for Claude Code, Codex, Cursor and VS Code; any other MCP client works with the configuration the same tab gives. Each answer an agent gets uses 1–10 credits.

See it on a real product

Eight models are published in full and open without an account. Ask one of them the question this page is about.