An MCP server that gives your coding agent the business logic
One line adds GitMir to Claude Code, Cursor, Codex or VS Code. From then on the agent asks the product — the rules, customer paths and features a task touches — before it edits code, instead of rebuilding them from files.
In short
MCP (Model Context Protocol) is the standard way coding agents call outside tools. Most MCP servers give an agent an action or a document — a browser, a database, the docs of a library. GitMir's MCP server gives it the business logic of your own product: what a task touches, what depends on it, which rules apply and which commit the answer is current to. Add it to Claude Code with one command — claude mcp add --transport http gitmir-vibe https://vibe.gitmir.com/mcp — or try a public model first, with no account and no key.
One line, then sign in once
The agent opens the browser once so you can sign in and choose the repository. There is no key to paste.
claude mcp add --transport http gitmir-vibe https://vibe.gitmir.com/mcpThen type /mcp in Claude Code and sign in.
codex mcp add gitmir-vibe --url https://vibe.gitmir.com/mcp
codex mcp login gitmir-vibe{
"mcpServers": {
"gitmir-vibe": {
"url": "https://vibe.gitmir.com/mcp"
}
}
}{
"servers": {
"gitmir-vibe": {
"type": "http",
"url": "https://vibe.gitmir.com/mcp"
}
}
}Cursor and VS Code also install it in one click from the GitMir Vibe page. Any other client that speaks MCP uses the same address.
Try it on a real product first
Every public GitMir model answers over MCP with no key. Point your agent at one and ask how that product works.
claude mcp add --transport http gitmir-supabase https://gitmir.com/mcp/supabaseAnswers about the product, not more files
Before a change
The rules, customer paths, features and records the task touches and how they connect — including parts that share no word with the task.
Checking a change
What the change touched, what depends on it but was left out, and which rules it reaches, before the commit lands.
What changed
What a commit changed in the product, and how far the model is from the agent's own branch.
What deserves a look
Findings about the product: actions with no recorded rule about who may take them, statuses nothing leads out of, risky exposure.
Work
Tasks a person accepted, with the parts of the product they touch and the check that says they are done.
Freshness
Every answer names the commit it is current to, so the agent never treats an old picture as the whole truth.
Each server answers a different question
| Question the agent has | Kind of MCP server | Example |
|---|---|---|
| How does this library's current API work? | Documentation | Context7 |
| What is in this repository, issue or pull request? | Code host | GitHub MCP server |
| Does the page actually work? | Browser | Playwright MCP |
| What is in the database right now? | Database or platform | Postgres or Supabase MCP servers |
| What does my product do, and what will this change break? | Business logic of your own product | GitMir |
The first four tell the agent about the world around your code. Only the last tells it what your product does.
Free to start, priced by what the agent asks
GitMir Vibe
One MCP server that checks each change before it is committed. Free to start with 10,000 credits, then $9 a person a month.
GitMir Intelligence
The model the server answers from — with the drawing, change impact and document checks for people in your GitMir account.
What a question costs
Each answer an agent gets uses 1 credit, 10 when GitMir picks the parts of the product a task touches. Questions from people are not metered.
Asked about MCP server
Run claude mcp add --transport http with a name and the server's address. For GitMir: claude mcp add --transport http gitmir-vibe https://vibe.gitmir.com/mcp — then type /mcp in Claude Code and sign in.
In .cursor/mcp.json inside the project, or ~/.cursor/mcp.json for every project. A remote server needs only its address under mcpServers; GitMir also installs in Cursor with one click from the GitMir Vibe page.
Search and retrieval servers return code that looks like the task; GitMir returns the connected business logic the task touches. In Benchmark 001 — one repository, fifty engineering questions, the same AI model and settings for every system — answers built on GitMir's model scored 92% against 78% for the strongest alternative, lost 8% of the score to errors and omissions against 22%, and used 8.4K tokens of context per question instead of 27.4K.
Yes. Every public GitMir model answers over MCP at https://gitmir.com/mcp/<model> with no key and no account — for example https://gitmir.com/mcp/supabase.
Yes. GitMir Vibe is com.gitmir/vibe (https://vibe.gitmir.com/mcp, sign-in by OAuth) and GitMir Intelligence is com.gitmir/intelligence (https://lab.gitmir.com/mcp, with the key from your GitMir account) at registry.modelcontextprotocol.io.
Add it to .vscode/mcp.json under servers, or run MCP: Add Server from the Command Palette; MCP servers need VS Code 1.102 or later and run in Chat's Agent mode. GitMir installs in one click from gitmir.com/with/vs-code.
GitMir reads the repositories you pick on GitHub, read-only, and builds the model from them. Nothing you connect is used to train models or shared with anyone.
It is free to start with 10,000 credits and no card. Each answer an agent gets uses 1 credit, or 10 when GitMir picks the parts a task touches; GitMir Vibe is $9 a person a month and adds a check of each change before it is committed.
Read next
CLAUDE.md and AGENTS.md from your code →Context engineering for coding agents →GitMir with Claude Code →GitMir with Cursor →GitMir with Codex →GitMir with VS Code →
More guides
CLAUDE.md and AGENTS.md →Context engineering →Architecture diagram from code →Change impact analysis →Spec-driven development →PRD template →Business rules extraction →All guides →
Give your AI the business logic of your product
Connect a repository and ask it in plain words — 10,000 credits to start, no card.