Skip to content

Add MCP HTTP proxy mode#315

Open
Lauritz-Timm wants to merge 1 commit into
rtk-ai:developfrom
Lauritz-Timm:fix/mcp-http-proxy
Open

Add MCP HTTP proxy mode#315
Lauritz-Timm wants to merge 1 commit into
rtk-ai:developfrom
Lauritz-Timm:fix/mcp-http-proxy

Conversation

@Lauritz-Timm

Copy link
Copy Markdown

Summary

  • add a lightweight icm serve --http-proxy <URL> MCP stdio proxy
  • expose POST /mcp on the warm HTTP daemon so multiple clients can share one store/embedder process
  • document the shared-daemon setup and RAM tradeoff for multi-client MCP users

Root cause

Each MCP stdio client launches its own icm serve process. With embeddings enabled, every process can load its own ONNX model, so idle RAM usage scales with the number of AI clients.

Validation

  • cargo check -p icm-cli --no-default-features --features http-api,backend-sqlite
  • cargo test -p icm-cli --bin icm proxy --no-default-features --features http-api,backend-sqlite,embeddings
  • cargo test -p icm-mcp --no-default-features --features backend-sqlite

Closes #314

@Lauritz-Timm Lauritz-Timm changed the base branch from main to develop July 2, 2026 13:00
@Lauritz-Timm Lauritz-Timm marked this pull request as ready for review July 2, 2026 13:01
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

MCP stdio servers load duplicate embedding models across AI clients

1 participant