Teo BouancheauBusiness Automation
Token Stunt
Smart code search MCP server for Claude Code. Parses source with tree-sitter, indexes every symbol and dependency edge into SQLite FTS5, and serves them through six MCP tools so agents read only what they need.
← All Projects2026
- Technologies
- Rust, tree-sitter, SQLite FTS5, Model Context Protocol, Cargo workspace, Ollama
- Live Site
- github.com/teobouancheau/tokenstunt
Business Automation2026
What it does
Token Stunt parses your source code with tree-sitter, extracts every function, class, interface, trait, enum, and constant, indexes them into a SQLite FTS5 store, and serves them through 6 MCP tools: ts_search, ts_symbol, ts_context, ts_overview, ts_setup, and ts_impact. Agents read only what they need instead of loading whole files.
Supported languages
Built-in: TypeScript, TSX, JavaScript, Python, Rust, Go, Java, C, C++, Ruby.
Optional (feature-gated): Swift, Kotlin, Dart.
Features
- Native Claude Code output: Unicode compact blocks, box-drawing characters, score bars, tree connectors, and aligned columns.
- Live reactivity: a filesystem watcher re-indexes changed files as you work.
- Dependency graph:
ts_contextshows callers and callees for any symbol. - Semantic search (optional): combine keyword ranking with embeddings via Ollama, LM Studio, or any OpenAI-compatible endpoint.
- Incremental indexing: only changed files are re-indexed on startup.
- Transparent storage: the index lives outside your project, so nothing is added to your repo.
Services
Frequently Asked Questions
- What problem does Token Stunt solve?
- It indexes an entire codebase with tree-sitter and SQLite FTS5 so AI coding agents can search and retrieve only the exact functions, classes, or symbols they need, instead of loading whole files into context.
- Which languages does Token Stunt support?
- TypeScript, TSX, JavaScript, Python, Rust, Go, Java, C, C++, and Ruby out of the box, with Swift, Kotlin, and Dart available as optional feature-gated support.
