MCP GitHub Project Manager
An MCP server that turns GitHub Projects v2 into a work queue for agents: they claim tasks, report progress and spend a budget.
- 105 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The server gets write access to GitHub: it creates and changes issues, pull requests, branches and releases, and the token opens repositories and organization projects.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/mcp-github-project-managerDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The server gets write access to GitHub: it creates and changes issues, pull requests, branches and releases, and the token opens repositories and organization projects.
This is third-party code. Review the repository files before installing.
What it does
MCP GitHub Project Manager is an MCP server that turns GitHub Projects v2 into a task management layer for AI agents. It exposes 21 compound tools and 175 actions, and storage is GitHub itself: issues, project fields and comments, with no database, Redis or external infrastructure. Agents register with a role and skills, claim tasks by priority, age or deadline, send heartbeats with progress and blockers and submit work for review: a branch, a pull request and test results. Each agent has its own token budget with warning thresholds and a hard stop. It also generates PRDs and parses them into issues, does triage, sprint planning and roadmaps, and works directly with pull requests, releases, branches and workflows. AI tools are enabled with model keys and are optional.
Who it is for. Teams where several agents, such as Claude Code and Codex, work on one repository and need a shared task queue without collisions.
Good fit when
- Agents pick the next task from the queue themselves and never pick the same one.
- A PRD is generated from an idea and split into issues with a complexity estimate.
- Each agent needs a token limit with a stop when it is exhausted.
- A sprint is planned from team capacity, priorities and risk based on the project's real issues.
Not a fit when
- There are few tasks and the GitHub web UI is enough: the server adds 175 actions you do not need.
- The repository is not on GitHub: it is the only storage.
- Agents must not get write access to projects and pull requests.
Example request
Register as an engineer agent, check out the highest-priority project task and show the issue context.MCP
- Transport
- stdio
- Authentication
- key is optional
Environment variables
GITHUB_TOKENsecret- GitHub token with repo, project and write:org scopes. A gh or OAuth sign-in can replace it.
GITHUB_OWNERrequired- Repository owner: a user or organization.
GITHUB_REPOrequired- Target repository name.
ANTHROPIC_API_KEYsecret- Model key for the optional AI tools. OpenAI, Google or Perplexity keys also work.
Limitations
Node.js 22.12 or newer is required. Access needs one sign-in method: a `GITHUB_TOKEN` with `repo`, `project` and `write:org` scopes, a GitHub App, the OAuth device flow or a `gh` login. `GITHUB_OWNER` and `GITHUB_REPO` are mandatory. Write tools create and change issues, pull requests, releases and branches, so grant the token only the permissions it needs. AI tools work only with model keys. Webhooks and SSE need a secret, otherwise request signatures are rejected.
How to disable. Remove the `github-project-manager` entry from the client configuration or run `claude mcp remove github-project-manager`. Uninstall the global package with `npm uninstall -g mcp-github-project-manager`.
FAQ
How do I install the server?
Run `npm install -g mcp-github-project-manager` or connect through `npx -y mcp-github-project-manager`. Claude Code has `claude mcp add github-project-manager -- npx -y mcp-github-project-manager`.
Is a GitHub token required?
Not necessarily. A `gh auth login` with the `project` scope or the OAuth device flow through `mcp-github-project-manager auth login` also works.
Why about twenty tools instead of a hundred?
Each tool takes an `action` string that selects the operation, so the client sees 21 schemas instead of 175.
How do I limit the tool set?
The `MCP_TOOL_GROUPS` variable keeps only the groups you need: `core`, `ai`, `agents`, `events` or `system`.
How do I control agent spending?
Each agent has a token budget with warning thresholds and a hard stop, and `record_usage` records the agent's own spend.
Discussion
No questions or reviews yet.
Related
MCP serversMCP server with a work item graph and server-side gates: an agent cannot start work until the required notes exist
A git MCP server with 28 tools: an agent clones, commits, branches, merges, rebases and manages worktrees in local repositories.
A remote MCP server that gives the agent docs and code of any public GitHub repository with nothing to install
An MCP server and CLI for code research: search and understand your project and all of GitHub with fewer tokens
A plugin with skills for building TypeScript MCP servers and MCP apps: the agent scaffolds, edits, debugs and verifies them
An MCP server for GitLab and self-hosted instances: projects, merge requests, issues, pipelines, wiki and releases
README in short
The repository holds a TypeScript server with 21 tools, six resources and six prompt templates, deployment and configuration documentation, skills for agents working on the project itself and a Docker image. A log of state changes can optionally be written to `.mcp-audit/mutations.jsonl`.