SETUP — How to Get This Build Working
1. Start Ollama (required for pipeline tools)
ollama serve # start Ollama
ollama pull qwen3.5:4b # pull the model (if not present)
Without Ollama running, 6 of 7 pipeline tools return empty results.
2. Restart OpenCode
Close and reopen OpenCode Desktop. The vibeserve MCP path has been fixed and context7 documentation lookup has been enabled.
3. Verify MCP servers
In OpenCode, check MCP status:
vibeserve ✓ connected (Python MCP server on localhost)
context7 ✓ connected (documentation lookup)
gh_grep ○ disabled (optional — enable for code search)
4. Run tests
python -m pytest tests/ -v --no-cov
5. Run a demo
python -m vibeserve --vibe-demo
This runs vibe_architect → vibe_code end-to-end. With Ollama running, the LLM calls will succeed and you'll get real output.
What was fixed
| Issue | Before | After |
|---|---|---|
| vibeserve MCP path | .../AetherNexus-MCP-main/vibeserve.py (deleted) |
.../VibeServe/vibeserve/core.py |
| PYTHONPATH | Missing | Set to VibeServe root |
| context7 MCP | Disabled | Enabled — real-time docs for React, FastMCP, Playwright |
| AGENTS.md | Outdated, referenced deleted project | Updated with real file map, honest status |
| Agents | 2 basic agents | Added build agent + security auditor |
| OPENCODE.md | Missing | Created with project context |