Skip to content

v0.12.0: MCP text resources

Latest

Choose a tag to compare

@he-yufeng he-yufeng released this 12 Nov 03:43
3a7d484

CoreCoder can now read text references published by MCP servers, over either stdio or Streamable HTTP. A server that advertises resources gets one on-demand bridge for its catalog, URI templates and text reads, including resource-only servers with no tools.

Cursors pass through unchanged and resource URIs stay with the configured server. Reads retain the existing consent, plan-mode and hook checks. Binary contents and results over 32,768 characters fail explicitly. This does not add retrieval indexing, subscriptions, dependencies or another branch to the agent loop.

The runtime change adds 69 physical lines to the MCP adapter. The engine remains 1,483 net lines; the package is 26 Python files and 3,276 physical lines. Both READMEs include examples and the source walkthroughs match the current implementation.

Validation covers Python 3.10 through 3.13 on Linux, macOS and Windows: 353 tests passed per job, with the opt-in live-model smoke skipped. The Linux package check also installed the built wheel, matched all 26 source files, exercised the console entry outside the checkout, and passed 42 resource protocol tests against the installed client. No external MCP service or commercial model was used. The publish workflow repeats the installed-wheel check before uploading.

Implemented in #58, closing #57.