Github Repo Intelligence MCP

What github-repo-intelligence-mcp MCP server does

The github-repo-intelligence-mcp MCP server turns GitHub repository data into a maintainability assessment. It accepts either an owner/name value or a GitHub repository URL and produces an overall verdict: Actively maintained, Slowing, At-risk, or Likely abandoned.

The assessment covers four dimensions:

• Activity, including commit cadence, releases, and momentum
• Issues, including response times and stale backlog share
• Pull requests, including merge rate and queue age
• Contributors, including contributor count and concentration risk

The headline response includes a per-dimension breakdown, a rationale identifying the weakest area, repository stars and forks, the fetch timestamp, and whether the request was authenticated. Raw values remain available so an agent or developer can inspect how the result was reached.

How it works

The github-repo-intelligence-mcp MCP server uses pinned thresholds to classify each dimension. Activity considers recent commits and can mark a repository Likely abandoned when it has had no commits for more than 365 days. Issue and pull request classifications use response, backlog, merge, and age measurements. Contributor analysis flags concentration when one contributor accounts for more than half of commits or the top two account for more than 80%.

The overall verdict is activity-dominant and evaluated in order. An abandoned activity signal takes precedence; other At-risk combinations can also produce an At-risk result. Activity that is healthy while another dimension is weak, or activity that is moderate, generally produces Slowing. Identical repository data and configuration produce the same result.

One GitHub fetch is cached and shared across the five tools, so dimension-specific follow-up calls use the same payload rather than making another upstream request.

Setup and configuration

Deploy the Actor on Apify in Standby mode. It starts without required startup input and exposes the Streamable HTTP MCP endpoint at /mcp. Configure an MCP client with an endpoint such as https://<your-standby-url>/mcp, then call a tool with a repo argument.

A GitHub token is effectively required in practice. The primary GitHub GraphQL request rejects unauthenticated access, and anonymous requests from Apify shared IP addresses can exhaust GitHub's quota quickly. Without authentication, the Actor returns rate-limit errors instead of repository verdicts. The token is supplied through the Actor's githubToken input field.

Tools and capabilities

The github-repo-intelligence-mcp MCP server provides five tools, each accepting one repo argument. get_repo_health returns the overall verdict and dimension summary. get_activity_metrics reports commit counts across 30-, 90-, and 365-day windows, release activity, and a momentum trend. get_issue_health reports open and closed issues, response and close-time medians, and stale issue ratios.

get_pr_health provides open, merged, and closed-unmerged pull request counts, merge rate, merge-time data, and the age of the oldest open PR. get_contributor_insights reports all-time contributors, the share held by the top contributors, and the bus-factor flag.

Some values can be null when there are no qualifying events, such as merge-rate data without decided pull requests. A repository that uses tags instead of GitHub Releases can show zero releases without that fact alone reducing its verdict.

Limitations and notes

The results measure observable GitHub activity and workflow signals; they are not a complete assessment of code quality, security, compatibility, or project governance. Contributor totals are all-time counts from GitHub's contributors list, not counts of recently active contributors. Stale issues are those untouched for more than 90 days. Merge rate excludes pull requests that remain open, and solo maintainers can receive no higher than a Moderate contributor classification.