Skip to content

stitcher-scout

LLM-powered GitHub code scout — finds real, working code relevant to your project.

Give it a description of what you want to build. It decomposes the problem into sub-problems, searches GitHub for implementations, reads actual source code to evaluate quality and relevance, and produces a structured report with recommended repositories and files.

Works with any LLM provider: OpenAI, Anthropic, Google Gemini, Ollama, and 100+ others via litellm.

Quick start

pip install stitcher-scout

stitcher setup    # interactive credential setup (or set env vars manually)

stitcher scout "A real-time multiplayer game server in Rust with WebSocket support"

See Getting Started for full setup instructions.

What it does

Description ──► Decompose ──► Search ──► Evaluate ──► Report
                                  ▲          │
                                  │          ▼
                                  └── Refine ◄── (deep mode only)
  1. Decompose — An LLM breaks your description into sub-problems (core libraries, architecture patterns, specific features)
  2. Search — Each sub-problem generates multiple GitHub queries with stratified search (by stars, recency, mid-range). Results are cached locally for speed.
  3. Evaluate — The LLM reads actual source code from candidate repos, scoring relevance and quality
  4. Deduplicate — Repos appearing across multiple sub-problems are consolidated; cross-cutting "Swiss Army knife" repos are flagged
  5. Refine (deep mode) — Extracts domain vocabulary from top results, follows dependency graphs, generates new searches
  6. Report — Produces a structured report with recommended repos, ecosystem map, patterns & insights, quality signals, and cost summary

See How It Works for the full pipeline breakdown.

Use as a CLI tool

# Quick search
stitcher scout "OAuth2 service with PKCE flow"

# Deep search with refinement
stitcher scout --mode deep "GPU cluster scheduler"

# Preview the search strategy before running
stitcher scout --explain "GPU cluster scheduler"

# Generate a research brief + dependency manifest
stitcher scout --brief "WebSocket server in Python"

# Use a different model
stitcher scout --model gpt-4o "Event sourcing in Go"

# Save report to file
stitcher scout -o report.md "WebSocket server in Python"

Use as an MCP tool in Claude Code

claude mcp add --scope user stitcher stitcher-mcp

After restarting Claude Code, the scout tool is available for AI-assisted project research. See MCP Integration.

Supported models

Provider Example Env var
Anthropic claude-sonnet-4-20250514 (default) ANTHROPIC_API_KEY
OpenAI gpt-4o OPENAI_API_KEY
Google Gemini gemini/gemini-2.0-flash GEMINI_API_KEY
Ollama (local) ollama/llama3 None
OpenRouter openrouter/anthropic/claude-3.5-sonnet OPENROUTER_API_KEY
Together AI together_ai/meta-llama/Llama-3-70b TOGETHER_API_KEY

See Configuration for all settings.

License

MIT