feat: serve over streamable HTTP with per-request credentials

[zvikagart]

Lets a single server process serve many users, so callers (notably the Code Ocean agent) can run one shared server instead of spawning one subprocess per invocation.

Changes

• New --transport streamable-http mode (with --host/--port), served with stateless_http=True. stdio remains the default; running the server with only CODEOCEAN_TOKEN and CODEOCEAN_DOMAIN exported from Claude Desktop or Cursor is unchanged.
• The CodeOcean client is now resolved per request in HTTP mode, from the request's Authorization: Bearer header, via a request-scoped proxy. Tools keep closing over a single client object, so all 26 tool definitions, their names, input schemas and descriptions are untouched.
• Tool descriptions are read at registration time, before any request exists; the proxy falls back to a placeholder client built from the environment for those reads, which makes no network call.
• HTTP requests without a usable credential are refused rather than falling back to a server-wide token.
• Per-token clients are cached so connection pools are reused instead of growing with invocation count.

[Copilot]

Pull request overview

Adds a new streamable-HTTP transport option so a single codeocean-mcp-server process can serve multiple callers, resolving Code Ocean credentials per request from the HTTP Authorization header while keeping the existing stdio behavior unchanged.

Changes:

• Introduces --transport streamable-http (plus --host / --port) and runs FastMCP in stateless HTTP mode.
• Adds a request-scoped CodeOcean client proxy with an LRU cache to reuse connection pools per token.
• Adds end-to-end tests for per-request credential isolation and updates README usage docs.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
tests/test_http_transport.py	Adds E2E coverage for streamable-HTTP, per-request tokens, and tool parity with stdio.
src/codeocean_mcp_server/server.py	Adds CLI parsing and wires streamable-HTTP transport + request-scoped client behavior.
src/codeocean_mcp_server/client.py	Implements per-request token extraction and cached client creation.
README.md	Documents how to run and call the server over streamable HTTP.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.