🌿 Gitrama MCP Server

AI-powered Git intelligence for your IDE — smart commits, branch names, PR descriptions, diffs, code review, push, and workflow management.

What is this?

Gitrama MCP Server exposes Gitrama's CLI as 15 MCP tools that any AI assistant can use. Instead of typing gtr commit in your terminal, your AI assistant calls the tool directly — analyzing your code changes, generating commit messages, suggesting branch names, reviewing code, and more.

Works with: Cursor · Claude Desktop · Claude Code · Windsurf · VS Code · Zed · any MCP-compatible client

---

What's new in v1.3.2

• 5 new tools — gitrama_scan, gitrama_diff, gitrama_review, gitrama_status, and gitrama_push
• Version surfaced from health check — gitrama_health now returns the MCP server version so you can confirm exactly what's running
• Interactive HTML diff panel — gitrama_diff launches a browser panel with risk-annotated diffs, churn rates, coupling context, and contributor info overlaid on every changed file
• Push from chat — gitrama_push supports upstream tracking, force-with-lease, and auto-resolves the current branch
• Remote MCP server — connect from anywhere via https://mcp.gitrama.ai/mcp, no local install required
• Tool count updated from 10 → 15

---

Install (< 60 seconds)

Step 1: Install the package

pip install gitrama-mcp

Or with uv:

uv pip install gitrama-mcp

This installs both the MCP server and the gitrama CLI.

Step 2: Connect to your IDE

<details> <summary><b>Cursor</b></summary>

Add to .cursor/mcp.json in your project (or global settings):

{
  "mcpServers": {
    "gitrama": {
      "command": "gitrama-mcp",
      "env": {
        "GITRAMA_TOKEN": "your_token_here"
      }
    }
  }
}

</details>

<details> <summary><b>Claude Desktop</b></summary>

Add to claude_desktop_config.json:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "gitrama": {
      "command": "gitrama-mcp",
      "env": {
        "GITRAMA_TOKEN": "your_token_here"
      }
    }
  }
}

</details>

<details> <summary><b>Claude Code</b></summary>

claude mcp add gitrama gitrama-mcp

</details>

<details> <summary><b>VS Code</b></summary>

Add to .vscode/mcp.json:

{
  "mcpServers": {
    "gitrama": {
      "command": "gitrama-mcp",
      "env": {
        "GITRAMA_TOKEN": "your_token_here"
      }
    }
  }
}

</details>

<details> <summary><b>Windsurf</b></summary>

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "gitrama": {
      "command": "gitrama-mcp",
      "env": {
        "GITRAMA_TOKEN": "your_token_here"
      }
    }
  }
}

</details>

<details> <summary><b>Zed</b></summary>

Add to Zed settings (⌘,):

{
  "context_servers": {
    "gitrama": {
      "command": {
        "path": "gitrama-mcp"
      },
      "env": {
        "GITRAMA_TOKEN": "your_token_here"
      }
    }
  }
}

</details>

<details> <summary><b>Remote — no local install required</b></summary>

Connect directly to the hosted Gitrama MCP server. Works with Claude.ai, Claude Desktop, Cursor, Windsurf, VS Code, and any MCP-compatible client:

{
  "mcpServers": {
    "gitrama": {
      "url": "https://mcp.gitrama.ai/mcp",
      "env": {
        "GITRAMA_TOKEN": "your_token_here"
      }
    }
  }
}

Or install via Smithery:

smithery mcp add GitramaLLC/gitrama-mcp

</details>

Step 3: Done.

Ask your AI: "Commit my staged changes" — and watch it call gitrama_commit.

---

Tools (15)

Health & Diagnostics

Tool	Description
gitrama_health	Check AI server health and confirm MCP server version

Repository Intelligence

Tool	Description
gitrama_ask	Ask any question about your repo — ownership, history, risk, changes
gitrama_scan	Full structural health scan — continuity risk, boundary entropy, recurrence patterns
gitrama_status	Show working tree status with AI interpretation

Code Review & Diff

Tool	Description
gitrama_diff	Risk-annotated diff with interactive HTML panel — churn, coupling, contributor context
gitrama_review	AI code review before you push — security, correctness, risk, coupling

Commit Intelligence

Tool	Description
gitrama_commit	Generate an AI commit message for staged changes
gitrama_stage_and_commit	Stage files + commit in one step
gitrama_unstage	Remove files from staging without discarding changes

Branch & Push Management

Tool	Description
gitrama_branch	Create a branch from a natural language description
gitrama_push	Push current branch to remote with upstream and force-with-lease support

PR & Changelog

Tool	Description
gitrama_pr	Generate a PR description from branch diff
gitrama_changelog	Generate a changelog between refs

Stream (Workflow) Management

Tool	Description
gitrama_stream_status	Show current workflow stream
gitrama_stream_switch	Switch to a different stream
gitrama_stream_list	List all streams in the repo

---

Tool Details

gitrama_health

Check AI server connectivity and confirm the running MCP server version.

Example prompt: "Run a gitrama health check"

Example output:

✅ AI server is healthy!
   🤖 Model: grok-4.20-reasoning
   🌐 Connected to: https://api.x.ai/v1

🔖 Gitrama MCP Server: v1.3.2

---

gitrama_scan

Run a full structural health scan of the repository. Scores every file for continuity risk, boundary entropy, and recurrence patterns. Results are cached in last_scan.json for use by gtr diff and gtr review.

Example prompt: "Run a full gitrama scan of my repo"

---

gitrama_diff

Show a risk-annotated diff of current changes. Opens an interactive HTML browser panel with Gitrama's structural intelligence overlaid — risk scores, churn rates, coupling gaps, and contributor context for every changed file.

Parameters:

Parameter	Type	Default	Description
target	string	""	Branch or commit to diff against
staged	bool	false	Diff staged changes only

Example prompts:

• "Show me a diff of my staged changes"
• "Diff my branch against main"

---

gitrama_review

Run an AI code review on current changes before committing. Returns severity-graded findings — security, correctness, risk, coupling — plus a verdict and suggested commit message.

Parameters:

Parameter	Type	Default	Description
mode	string	"staged"	"staged", "uncommitted", "quick", "full"

Example prompts:

• "Review my staged changes"
• "Do a full review of all my uncommitted changes"

---

gitrama_push

Push the current branch to a remote repository. Uses --force-with-lease for safe force pushes and auto-resolves the current branch if none is specified.

Parameters:

Parameter	Type	Default	Description
remote	string	"origin"	Remote to push to
branch	string	""	Branch to push (default: current branch)
force	bool	false	Force push with --force-with-lease
set_upstream	bool	false	Set upstream tracking branch (-u)

Example prompts:

• "Push my changes"
• "Push this branch and set upstream"
• "Force push the current branch"

---

gitrama_status

Show the working tree status with AI interpretation of staged, unstaged, and untracked files.

Example prompt: "What's my current git status?"

---

gitrama_ask

Ask a natural language question about your repository. Gitrama analyzes commit history, file structure, blame data, and diffs to answer.

Parameters:

Parameter	Type	Default	Description
question	string	required	Any question about your repo
stream	string	""	Optional stream context override
deep	bool	false	Enable full repo history access

Example prompts:

• "Who owns the auth module?"
• "What's the riskiest file in this repo?"
• "What changed in the last 3 days?"
• "Explain the purpose of src/utils/retry.py"

---

gitrama_commit

Generate an AI-powered commit message for staged changes.

Parameters:

Parameter	Type	Default	Description
message	string	""	Optional custom message (skips AI generation)

Example prompt: "Commit my staged changes"

---

gitrama_stage_and_commit

Stage files and commit in one step.

Parameters:

Parameter	Type	Default	Description
files	string	"."	Files to stage (. for all, or space-separated paths)
message	string	""	Optional custom message

Example prompt: "Stage and commit all my changes"

---

gitrama_unstage

Remove files from the staging area without discarding changes.

Parameters:

Parameter	Type	Default	Description
files	string	""	Space-separated file paths to unstage
all_files	bool	false	Unstage everything currently staged

Example prompt: "Unstage src/auth.py"

---

gitrama_branch

Generate an AI-powered branch name from a description and optionally create it.

Parameters:

Parameter	Type	Default	Description
description	string	required	What you're working on
create	bool	true	Create and switch to the branch

Example prompts:

• "Create a branch for adding OAuth2 support"
• "Suggest a branch name for fixing the payment timeout, don't create it"

---

gitrama_pr

Generate a PR description from the diff between the current branch and base.

Parameters:

Parameter	Type	Default	Description
base	string	""	Target branch (default: main/master)

Example prompt: "Write a PR description for this branch"

---

gitrama_changelog

Generate a changelog between refs.

Parameters:

Parameter	Type	Default	Description
since	string	""	Start ref (tag, branch, hash)
until	string	""	End ref (default: HEAD)
format	string	"markdown"	"markdown" or "json"

Example prompt: "Generate a changelog since v1.1.4"

---

Stream Tools

Tool	Parameters	Example prompt
gitrama_stream_status	none	"What stream am I on?"
gitrama_stream_switch	name, description	"Switch to the hotfix stream"
gitrama_stream_list	none	"List all my gitrama streams"

---

The v1.3.2 Workflow

With all 15 tools connected, your full dev loop runs from chat:

describe intent  →  stream switch
write code
ask gitrama what changed  →  diff (HTML panel)
review before push
commit with AI message
push
PR description generated

No terminal. No manual git commands.

---

Configuration

Environment Variables

Variable	Default	Description
GITRAMA_TOKEN	""	Your Gitrama API key
GTR_CWD	os.getcwd()	Working directory for git operations
GTR_MCP_TRANSPORT	"stdio"	Transport: "stdio" or "streamable-http"
GTR_MCP_HOST	"0.0.0.0"	HTTP host (when using streamable-http)
GTR_MCP_PORT	"8765"	HTTP port (when using streamable-http)

HTTP Transport (for CI/CD)

GTR_MCP_TRANSPORT=streamable-http GTR_MCP_PORT=8765 gitrama-mcp

Then connect your client to http://localhost:8765/mcp.

---

Requirements

• Python 3.10+
• Git installed and in PATH
• A Gitrama API key or local Ollama instance

Set your API key:

gtr config --key YOUR_API_KEY

Or use a local model:

gtr config --provider ollama --model llama3

---

Development

git clone https://github.com/GitramaLLC/gitrama-mcp.git
cd gitrama-mcp
pip install -e ".[dev]"

# Test with MCP Inspector
mcp dev src/gitrama_mcp/server.py

---

License

Proprietary — see LICENSE.

---

Built by Alfonso Harding · gitrama.ai

🌿