gitea-mcp
A local MCP server that connects Claude directly to your Gitea instance, enabling management of repositories, issues, pull requests, files, and commit history.
README
<div align="center">
<img src="assets/banner.png" alt="Gitea MCP" width="100%" />
Gitea MCP
A local MCP server that gives Claude Desktop — and any MCP client — the ability to work with a self-hosted Gitea instance.
</div>
MCP clients — Claude Desktop included — have no built-in way to reach a self-hosted Gitea instance. gitea_mcp closes that gap: it's a local MCP server that gives any MCP-compatible client the ability to list and search repos, manage issues and pull requests, and read/write files directly against your own Gitea instance.
As a bonus, it does this without routing through a third-party gateway — most MCP setups for Git hosting (Composio and similar) put a gateway between your agent and your infrastructure, holding your token and brokering every API call. gitea_mcp skips that entirely: it runs as a subprocess on your own machine, talks over stdio, and speaks directly to your Gitea instance — your token and repo data never pass through anyone else's server.
[!NOTE] This project is intentionally local-only. There's no hosted endpoint, no third-party account, and no server to deploy — just a command your MCP client launches on demand.
Features
- Direct connection — talks straight to
GITEA_BASE_URL, nothing in between - Repositories — list and search
- Issues — list (filtered by state) and create
- Pull requests — list (filtered by state) and create
- Files — read content and write (create or update) as a single commit, with SHA-based conflict protection
- Commits — list history, optionally scoped to one file
- Actionable errors — auth, not-found, permission, conflict, and validation failures all come back as specific, readable messages instead of raw exceptions
Prerequisites
- Python 3.9 or later
- pipx or uv (for the recommended install path below)
- A running Gitea instance you have API access to
- A Gitea personal access token
Install
Recommended — installs a gitea-mcp command directly from this repo, no manual venv management:
pipx install git+https://github.com/datamathcode/gitea-mcp.git
# or
uv tool install git+https://github.com/datamathcode/gitea-mcp.git
Manual / development install — for contributing, or an editable local install:
git clone https://github.com/datamathcode/gitea-mcp.git
cd gitea-mcp
python3 -m venv venv
source venv/bin/activate # venv\Scripts\activate on Windows
pip install -e .
Both paths register the same gitea-mcp command.
[!IMPORTANT] Dependencies are pinned to
mcp==1.9.4deliberately. Newermcpreleases (2.0.0+) restructured the package and no longer exposeFastMCPat the path this server imports it from.
Configure
Generate a token in Gitea under Settings → Applications → Generate New Token, scoped narrowly to what this server needs (repo read/write, issue read/write) rather than every permission available.
export GITEA_BASE_URL="https://gitea.example.com"
export GITEA_TOKEN="<your token>"
The server refuses to start if either variable is missing — no silent fallback to a hardcoded value.
Usage
gitea_mcp is a standard MCP server communicating over stdio — it works with any MCP-compatible client, not just Claude Desktop. Any client that can launch a local subprocess and speak the MCP protocol can use it; point it at the installed command's path and set GITEA_BASE_URL/GITEA_TOKEN in the environment it launches with.
Run standalone
gitea-mcp
(Running from a source checkout without installing still works too: python3 gitea_mcp.py.)
Add to an MCP client
gitea_mcp works the same way with every client below: point it at the installed command's absolute path, and set GITEA_BASE_URL/GITEA_TOKEN in the environment it launches with. Each client just has its own config file and format for expressing that.
First, find the installed command's absolute path:
which gitea-mcp
Claude Desktop
In Claude Desktop's MCP config (claude_desktop_config.json):
{
"mcpServers": {
"gitea": {
"command": "/absolute/path/from/which/gitea-mcp",
"env": {
"GITEA_BASE_URL": "https://gitea.example.com",
"GITEA_TOKEN": "<your token>"
}
}
}
}
[!IMPORTANT] Use the full path from
which gitea-mcp, not just"gitea-mcp". Claude Desktop, launched via Finder/Dock on macOS, doesn't inherit your shell'sPATHadditions — a bare command name resolves fine from a terminal but fails to launch from Claude Desktop's config. Check whether your client has the same limitation before assuming a bare command name will work.
Claude Code
claude mcp add --transport stdio gitea --env GITEA_BASE_URL=https://gitea.example.com --env GITEA_TOKEN=YOUR_TOKEN_HERE -- /absolute/path/from/which/gitea-mcp
Docs: code.claude.com/docs/en/mcp-quickstart
Codex
codex mcp add gitea --env GITEA_BASE_URL=https://gitea.example.com --env GITEA_TOKEN=YOUR_TOKEN_HERE -- /absolute/path/from/which/gitea-mcp
Docs: learn.chatgpt.com/docs/extend/mcp
Hermes Agent
In ~/.hermes/config.yaml:
mcp_servers:
gitea:
command: "/absolute/path/from/which/gitea-mcp"
env:
GITEA_BASE_URL: "https://gitea.example.com"
GITEA_TOKEN: "<your token>"
Docs: hermes-agent.nousresearch.com/docs/user-guide/features/mcp
OpenCode
In .opencode.json — note env here is an array of "KEY=value" strings, not an object like the other clients above:
{
"mcpServers": {
"gitea": {
"type": "stdio",
"command": "/absolute/path/from/which/gitea-mcp",
"args": [],
"env": ["GITEA_BASE_URL=https://gitea.example.com", "GITEA_TOKEN=<your token>"]
}
}
}
Docs: opencode-ai-opencode.mintlify.app/features/mcp-integration
Any MCP client launches the server the same way: as a subprocess, talking over stdio — no port to open, no service to keep running.
Available tools
| Tool | Description |
|---|---|
gitea_list_repos |
List or search repositories accessible to the authenticated user |
gitea_list_issues |
List issues in a repository, filtered by state |
gitea_create_issue |
Create an issue with a title, body, and existing labels |
gitea_list_pull_requests |
List pull requests in a repository, filtered by state |
gitea_create_pull_request |
Open a pull request from one branch into another |
gitea_get_file_content |
Read a file's content and current blob SHA |
gitea_create_or_update_file |
Create or update a file as a single commit |
gitea_list_commits |
List commit history, optionally scoped to a file path |
Scope
This is a first pass covering the core workflow — repos, issues, PRs, file read/write, and commit history. Branch management, releases, label administration, webhooks, and org-level operations aren't implemented yet. The plan is to expand once this core set proves out in real use rather than building everything speculatively up front.
Testing
tests/test_gitea_mcp.py exercises all 8 tools against a real Gitea instance — no mocking of the Gitea API.
[!WARNING] Point this at a disposable scratch repo, never a real project repo. The write tests create and clean up issues, pull requests, branches, and files as part of running.
pip install -e ".[dev]"
export GITEA_BASE_URL="https://gitea.example.com"
export GITEA_TOKEN="<your token>"
export TEST_REPO_OWNER="<owner of your scratch repo>" # defaults to "dmc"
export TEST_REPO_NAME="<name of your scratch repo>" # defaults to "gitea-mcp-test-repo"
pytest tests/ -v
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.