Support Streamable HTTP for MCP & Provide Docker

[msantosh-amdocs]

I run graphify on my code base as part of CI i generated graph.json.
Now i want to run a streamable HTTP server for MCP using generated graph.json from central location and support oauth as authentication or support api key.

Thisway i can ask my team to use latest mcp of codebase to understand what was written

[safishamsi]

Great use case — running a shared MCP server so the whole team queries the same graph is exactly the kind of deployment model we want to support.

Current state: The graphify MCP server (graphify/serve.py) runs stdio-only today. It is designed for the per-developer MCP model where each Claude Code / IDE session spawns its own local process connected to a local graph.json.

What Streamable HTTP would enable: A single server process in CI or on a shared host serving the graph over HTTP+SSE (the MCP Streamable HTTP transport introduced in MCP spec 2025-03-26), so any team member can point their IDE at the central URL without running graphify locally. Auth (API key or OAuth) lives at the HTTP layer.

What needs building:

1. An HTTP transport layer using mcp.server.fastmcp or starlette/uvicorn alongside the existing stdio path
2. A --transport http --host 0.0.0.0 --port 8080 flag on python -m graphify.serve
3. Optional --api-key / OAuth middleware for the HTTP path
4. A Docker image (graphify serve graphify-out/graph.json --transport http) so CI can publish the graph as a service

For now, a workaround: You can expose the stdio server over HTTP today using mcp-proxy or similar bridge tools that wrap a stdio MCP server in an HTTP/SSE transport. Not ideal but gets you unblocked.

Streamable HTTP transport is on the roadmap. A PR implementing the --transport http flag on graphify.serve would be very welcome — the MCP Python SDK already has the Streamable HTTP server primitives available.

[safishamsi]

It's up in 0.8.34, give it a shot and lmk how it goes

[msantosh-amdocs]

Thank you for adopting the feature faster !!
I placed it in my pipeline and was able to generate it and use will share feedback once my whole team starts using it