majiayu000 Intelligent Code Search with WarpGrep
WarpGrep is an RL-trained search agent that reasons about code, not just pattern matches.
How It Works
- 8 parallel searches per turn (explores multiple hypotheses)
- 4 reasoning turns (follows causal chains across files)
- F1=0.73 in ~3.8 steps (vs 12.4 for standard search)
When to Use WarpGrep
| Use WarpGrep | Use Built-in Grep |
|---|---|
| "How does auth flow work?" | "Find class UserService" |
| "Trace data from API to DB" | Simple regex patterns |
| "Find all error handling" | "Where is X defined?" |
| Large repos (1000+ files) | Known file patterns |
| Before major refactoring | Quick needle lookups |
Query Formulation
Good queries (reasoning required):
"How does authentication flow from the login handler to the database?"
"Find all places where user permissions are checked"
"Trace the request lifecycle from router to response"
Bad queries (use Grep instead):
"Find UserService" → use Grep
"Search for 'import React'" → use Grep
Workflow
- Formulate query: Describe WHAT you want to understand, not just WHAT to find
- Run WarpGrep:
mcp__morphllm__warpgrep_codebase_search - Interpret results: Ranked snippets with file paths and line numbers
- Follow up: Read specific files for deeper understanding
Parameters
search_string: "natural language description of what to find"
repo_path: "/absolute/path/to/repo"
Tips
- Be specific about the behavior or flow you're investigating
- Include context: "in the API layer" or "during startup"
- WarpGrep handles ambiguity better than exact pattern matching
- Results include surrounding context for understanding