Set up MCP bridge for Qwen3 via Ollama #1067

Closed
opened 2026-03-23 12:52:49 +00:00 by perplexity · 0 comments
Collaborator

Parent: #1063
Related: #982 (session crystallization tech stack)

Objective

Establish MCP (Model Context Protocol) integration so Qwen3-14B running in Ollama can call Timmy's MCP tools (Gitea, file system, shell, etc.).

Options (evaluate in order)

  1. Qwen-Agent native MCPpip install qwen-agent[mcp] — first-party support from Qwen team
  2. ollmcp — Ollama-MCP bridge
  3. mcphost — Generic MCP host that works with Ollama
  4. ollama-mcp-bridge — Purpose-built bridge

Steps

  1. Evaluate each bridge option for stability and feature completeness
  2. Install and configure the selected bridge
  3. Register Timmy's MCP tools (Gitea API, file operations, shell execution)
  4. Test tool calling end-to-end: prompt → model → MCP tool → result → model response
  5. Document the setup in docs/mcp-setup.md

Acceptance Criteria

  • Qwen3-14B can invoke MCP tools via structured JSON tool calls
  • Round-trip latency (prompt → tool call → response) under 5 seconds for simple tools
  • At least Gitea list/create issue tools working via MCP

Notes

Qwen3 models are explicitly listed among recommended Ollama models for MCP/tool use. The 32K context window is sufficient for most agent loops with tool response injection.

Parent: #1063 Related: #982 (session crystallization tech stack) ## Objective Establish MCP (Model Context Protocol) integration so Qwen3-14B running in Ollama can call Timmy's MCP tools (Gitea, file system, shell, etc.). ## Options (evaluate in order) 1. **Qwen-Agent native MCP** — `pip install qwen-agent[mcp]` — first-party support from Qwen team 2. **ollmcp** — Ollama-MCP bridge 3. **mcphost** — Generic MCP host that works with Ollama 4. **ollama-mcp-bridge** — Purpose-built bridge ## Steps 1. Evaluate each bridge option for stability and feature completeness 2. Install and configure the selected bridge 3. Register Timmy's MCP tools (Gitea API, file operations, shell execution) 4. Test tool calling end-to-end: prompt → model → MCP tool → result → model response 5. Document the setup in `docs/mcp-setup.md` ## Acceptance Criteria - Qwen3-14B can invoke MCP tools via structured JSON tool calls - Round-trip latency (prompt → tool call → response) under 5 seconds for simple tools - At least Gitea list/create issue tools working via MCP ## Notes Qwen3 models are explicitly listed among recommended Ollama models for MCP/tool use. The 32K context window is sufficient for most agent loops with tool response injection.
claude self-assigned this 2026-03-23 13:44:34 +00:00
claude added the harnessinferencep0-critical labels 2026-03-23 13:53:02 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Rockachopa/Timmy-time-dashboard#1067