GitHub Repository Intelligence
Repo intel for AI coding agents: overview, PRs, contributors, hot files, CI, deps. Remote MCP.
Open source Repository Open in the app JSON README (API)
About
Repo intel for AI coding agents: overview, PRs, contributors, hot files, CI, deps. Remote MCP.
Details
- Kind
- MCP servers
- Topic
- Version control
- Publisher
- heisencodex-jpg
- Origin
- official
- Category
- ferramentas
- Transport
- http
- Version
- 0.3.1
- Last push
- 2026-05-21T08:26:02Z
- Repository state
- ativo
- Language
- TypeScript
- Added
- 2026-08-29 04:00:07
- Updated
- 2026-08-29 04:00:07
- Origin id
io.github.heisencodex-jpg/github-repo-intel-mcp
README
# GitHub Repository Intelligence MCP Server
A hosted [Model Context Protocol](https://modelcontextprotocol.io) server that gives AI coding agents structured intelligence about any GitHub repository — overview, recent PRs, contributors, hot files, CI status, and dependencies — through a single MCP endpoint.
**No local setup. No `git clone`. No Octokit boilerplate.** Plug your agent into the MCP URL and start asking questions.
## Why this exists
Local GitHub MCP servers exist, but they require every user to set up Node/Python, install the server, manage tokens, and keep it patched. This Actor runs on Apify's platform and is reachable as a hosted streamable-HTTP MCP endpoint — your agent just needs the URL.
It's useful when you want an LLM to reason about a codebase without giving it filesystem or shell access: think code-review bots, repo-onboarding helpers, dependency-audit agents, refactor planners.
### Hosted vs local — when this saves you time
| | This Actor | Local MCP server |
|---|---|---|
| Setup | Paste URL + token into client config | `git clone`, install deps, manage Node/Python versions |
| Distribution | Same URL works for every teammate | Each teammate sets up their own |
| Updates | Always latest server build | You patch + redeploy |
| Auth surface | One Apify token + optional GitHub PAT in tool args | GitHub PAT lives on every machine |
| Cost when idle | $0 (Standby mode, billed per call) | Whatever your laptop/server costs |
| When local wins | Air-gapped repos, on-prem GitHub Enterprise | — |
### Three workflows it shortens
1. **Repo onboarding** — drop the URL into Claude Code, ask *"give me a one-page brief on `kubernetes/kubernetes`: maintainers, top 10 hot files, current PR pipeline, CI health."* One conversation, six tool calls, no clones.
2. **PR-review prep** — before reviewing a stranger's PR, run `get_contributors` + `list_recent_prs` + `get_hot_files` on the target repo to know who you're dealing with and where the churn actually lives.
3. **Dependency / supply-chain audit** — `get_dependencies` across a list of repos in a loop, no shell access required, no risk of malicious `npm install` running on your machine.
## Tools exposed
| Tool | What it returns | Typical use |
|---|---|---|
| `get_repo_overview` | Stars, forks, language, license, topics, last push | "Is this repo still maintained?" |
| `list_recent_prs` | Recent PRs with author, state, merge dates | "Show me the last 20 merged PRs by user X" |
| `get_contributors` | Top contributors by commit count | "Who maintains this project?" |
| `get_hot_files` | Most-modified files over the last N commits | "Where's the churn?" |
| `get_ci_status` | Check runs and conclusions for a ref | "Is the default branch green?" |
| `get_dependencies` | Parses `package.json`, `requirements.txt`, `go.mod`, `Cargo.toml`, `pyproject.toml` | "What does this project depend on?" |
Every tool accepts an optional `github_token` parameter. With no token you get GitHub's 60-req/hour limit on public repos. Provide a fine-grained PAT and you get 5,000 req/hour and can access private repos.
## Pricing
**$0.005 per tool call. First 100 calls free per user.**
Pay-per-event model — you only pay when a tool actually executes. No subscription, no minimums. The server runs in Apify Standby mode, so you're not billed for idle time.
## Usage
### From an MCP-aware client (Claude Desktop, Cursor, Continue, etc.)
The Actor exposes a Streamable-HTTP MCP transport at:
```
https://cg-nguyen--github-repo-intel-mcp.apify.actor/mcp
```
Authentication uses your Apify token via the `Authorization: Bearer <APIFY_TOKEN>` header — most MCP clients let you configure this in their server config.
### Example tool call (raw JSON-RPC)
```bash
curl -X POST https://cg-nguyen--github-repo-intel-mcp.apify.actor/mcp \
-H "Authorization: Bearer <APIFY_TOKEN>" \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "get_repo_overview",
"arguments": {"owner": "facebook", "repo": "react"}
}
}'
```
### From Claude Desktop config
```json
{
"mcpServers": {
"github-repo-intel": {
"url": "https://cg-nguyen--github-repo-intel-mcp.apify.actor/mcp",
"headers": {
"Authorization": "Bearer <APIFY_TOKEN>"
}
}
}
}
```
## Limits and caveats
- Free tier on Apify gets you $5/month of usage credits — enough for ~1,000 free tool calls before the per-event price kicks in.
- The `get_hot_files` tool scans up to 200 commits and makes one API call per commit, so it eats GitHub rate limit the fastest. Pass a `github_token` if you call it often.
- `get_ci_status` reads the GitHub **Checks API** — it surfaces GitHub Actions, CodeQL, and any third-party check that posts back via the Checks API. Repos that run their CI entirely on **external systems** (Prow, Buildkite, CircleCI without GitHub integration) may show 0 check runs even when the build is green. The response includes a `note` field flagging this case.
- This is a **read-only** server. It cannot create PRs, comment on issues, or modify any repository state.
- All data comes straight from the GitHub REST API — no caching layer, so freshness equals what the GitHub API returns.
## Source & roadmap
Author: [luongnp](https://github.com/luongnp). Open to feature requests — file an issue on the Actor page if you want a tool added (issues board, releases, dependency vulnerabilities via Dependabot API, etc.).