Adaptiv MCP (GitHub + Render)

Adaptiv MCP is a self-hosted, connector-oriented Model Context Protocol (MCP) server that exposes GitHub, Git and Render automation as MCP tools, plus a small HTTP registry/UI surface for discovery and debugging.

It’s designed to work cleanly with ChatGPT / OpenAI MCP connectors (including common preflight/probe behavior).

What you get

• You can use this App to use ChatGPT to find/update bugs, issues, tests, PR etc.
• You can use this App to create brand new Apps etc. without any knowledge of coding to begin with.
• You can use this App as a Professional Developer who uses ChatGPT in everyday workflows.
• You can FULLY customize this APP by modifying the code and adding new MCP tools for new endpoints for different uses other than just Github/Render/Coding etc.

Tooling

• GitHub automation
  • repos, branches, files
  • issues + comments
  • pull requests (open/merge/close/comment)
  • GitHub Actions workflows (list runs, trigger dispatch, fetch logs)
• Render automation
  • list services / deploys
  • trigger/cancel/rollback deploys
  • restart services
  • fetch logs
• Workspace / repo mirror (server-side git working copy)
  • apply file edits safely
  • run commands/tests in a real checkout
  • commit/push and open PRs via a “mirror-first” workflow

Connector-friendly server behaviors

• Two MCP transports
  • Streamable HTTP (preferred): POST /mcp
  • Legacy SSE transport: GET /sse + POST /messages
• Preflight-safe endpoints for OPTIONS / HEAD / GET probes
• Base-path safe links via GET /ui.json when deployed behind a reverse proxy prefix

Endpoints

MCP transports:

• Streamable HTTP: POST /mcp
• SSE: GET /sse and POST /messages

HTTP registry/diagnostics:

• GET /healthz – runtime health
• GET /tools – tool discovery used by connectors
• POST /tools/<tool_name> – invoke a tool over HTTP
• GET /resources – resource discovery
• GET /ui – lightweight UI (links + diagnostics)
• GET /ui/tools – tool catalog UI
• GET /ui.json – machine-readable service metadata (base-path aware)

Deploying on Render.com

1. Make a Render.com account and start a Web Service.

2. Use these Start/Build Commands

3. Setup Evironment Variables Example:

(ADAPTIV_MCP_CONTROLLER_REPO set to your mainly worked on Repo. can still see all repos scoped in the Github PAT.)

4. Open ChatGPT and go to Settings>Apps>Create App and set it up with the URL you recieved with /sse at the end.

5. Connect your chat message in ChatGPT with your custom made App.

6. You can now Clone any Repo and use with the provided Github OAUTH scope which you provide and enter into the environment variables on Render.com

Examples:

• Tell ChatGPT Assistant to use validate_environment to make sure everything is setup correctly.

GitHub auth

Set at least one token variable:

• GITHUB_TOKEN (common)
• GITHUB_PAT, GH_TOKEN, GITHUB_OAUTH_TOKEN (alternates)

Optional:

• GITHUB_API_BASE – override for GitHub Enterprise / custom API base (defaults to https://api.github.com)

Render auth

Set one token variable:

• RENDER_API_KEY (common)
• RENDER_API_TOKEN, RENDER_TOKEN (alternates)

Workspace / repo mirror

Workspace tools operate on a persistent server-side repo mirror (git working copy). Useful settings:

• MCP_WORKSPACE_BASE_DIR – where mirrors are stored (defaults under your cache directory)

Deployment notes

Render

This repo includes Render-friendly scripts:

• scripts/render-build.sh – installs dependencies with a hash marker to speed subsequent deploys
• scripts/render-start.sh – validates a vendored rg binary, normalizes log level for uvicorn, and starts the server

The start script expects a working rg on $PATH and will prefer the vendored binaries under vendor/rg/....