Decoder – GPS Navigation for Codebases
Call graph traversal for LLMs cuts token usage vs. iterative grep and file reads.
17 MCP query tools for codebase navigation — functions, classes, imports, dependency graphs, change impact. Zero dependencies. 87% token reduction.
MCP query server cuts codebase context by 87% with zero dependencies, measured.
Backend developers, full-stack engineers using Claude Code or Cursor
Sourcegraph Cody · Continue.dev · Cursor's symbol indexing
mcp-codebase-index parses your codebase into functions, classes, imports, and dependency graphs, then exposes 17 query tools via MCP.
Measured results: 58-99% token reduction per query (87% average). In multi-turn conversations, 97%+ cumulative savings.
Zero dependencies (stdlib ast + regex). Works with Claude Code, Cursor, and any MCP client.
pip install "mcp-codebase-index[mcp]"
Call graph traversal for LLMs cuts token usage vs. iterative grep and file reads.
AST + embeddings for codebase search—but Sourcegraph Cody, Cursor, and Continue already solve this.
AST-based code indexing saves tokens but Cursor and Continue already do this.
Tree-sitter + FTS5 + MCP = tokens saved for AI agents to actually code, not search.
Honest benchmark shows RAG overhead on trivial queries; 63% token savings on complex tasks.
Compressed JSON bundles fit tight context windows better than pasting files.