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
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。