AUTOMATION
Token-Efficient Code Search for AI Agents
Use Semble MCP server to search code with 98% fewer tokens than grep
Updated: 5/18/2026
Difficulty
easy
Time
5m
Use Case
Enable AI agents to efficiently search large codebases without token waste or subagent spawning
Popularity
0 views
About this automation
Integrate Semble MCP server into Claude Code or other agent platforms to provide token-efficient code search using static embeddings and BM25 fusion. Achieves 99% retrieval quality of transformer-based search while using 98% fewer tokens and running 200x faster on CPU.
How to implement
1
Install Semble via: claude mcp add semble -s user -- uvx --from 'semble[mcp]' semble
2
Configure MCP server in Claude Code settings
3
Query codebase using natural language or code patterns
4
Agent receives ranked results without full file reads
5
Reduce token consumption and improve search accuracy