#tree-sitter

Structural graph memory for AI coding assistants. repo-graph maps your codebase entities, relationships, and feature flows so the model navigates to the right files instead of reading everything first. Tree-sitter, 20+ languages, frontend to backend and more, any MCP client.

GraphPilot gives coding agents persistent structural knowledge of a codebase. It parses TypeScript/JavaScript with tree-sitter into a graph of symbols and call edges, then exposes that graph over MCP so the agent can recall definitions, find callers, and trace change impact without re-reading files every conversation. Local-first: no network, no telemetry, no code ever leaves your machine. Four tools — gp_index, gp_recall, gp_callers, gp_impact.

Deterministic cross-branch conflict detection for AI coding agents. Scans your git repo for function signature mismatches, HTTP route conflicts, and schema breaks across branches using tree-sitter AST analysis. Catches the merge failures that CI misses because tests only run on one branch at a time.

Ganglia indexes your project into a call graph and exposes 73 semantic tools — code_get, code_callers, code_impact, code_propagate, code_apply_batch with one-click revert, code_safe_move, deliberation chain, smart LLM-backed digests, cross-session memory, and more. Works with Claude Code, Cursor, Windsurf, Continue, Cline, Goose. Languages: TypeScript, JavaScript, Rust, Python, Go, PHP. Free during public beta. Token savings: 80–95% vs raw file-reading patterns.

MCP server that detects semantic (non-textual) merge conflicts between Git branches using AST-level analysis. Catches incompatible changes that Git merges cleanly but break at runtime — like signature changes, removed exports, interface breaks, and cross-file dependency conflicts.

Code-intelligence MCP server built in Rust. Pre-computes a knowledge graph (tree-sitter + SQLite + PageRank + blast radius + git co-change + cyclomatic complexity) and serves it to any AI agent. Replaces grep/read/find with 21 structural tools. Saves 91.5% tokens, 7.9/10 LLM-judge quality vs 5.3/10 without. Supports Claude Code, Cursor, Windsurf, Zed, Continue.dev, OpenCode, Codex CLI. Local-first, no cloud, no embeddings. `make deploy` and done. Note: config assumes `qartez-mcp` is already installed via `make deploy` (which puts it in `~/.local/bin/`). If the form requires a full path variant:

Save tokens while coding — your AI agent gets structured code context, not file dumps. CodeWeave is an MCP server that gives AI agents structured code intelligence instead of dumping entire files into context. Your agent queries local indexes — AST, call graph, type graph, and hybrid semantic search — and gets back only what it needs. The result: less token waste, more relevant context, better coding decisions. It supports 7 languages (Python, TypeScript, JavaScript, Go, Rust, Java, C#) with a 6-stage hybrid search pipeline combining vector embeddings, full-text search, and structural density scoring. Everything runs locally with zero external dependencies beyond Ollama for embeddings. Monorepo and git worktree support included. A file watcher keeps indexes current as code changes — no manual reindexing needed. ATTENTION: Run "npx @codeweave/mcp" in your project directory first. The setup wizard handles everything for you. At first, it may take some time. INFO: Exhaustively tested on large Java, C#, JavaScript, TypeScript, React codebases. Contributions and feedback are welcome.

MCP server that extracts decision intent from git history and protects intentional code from AI modification.

Stop re-explaining your codebase to AI every session. CogniLayer gives Claude Code persistent memory across sessions, AST-based code intelligence (who calls what, blast radius analysis), and subagent context compression (40K → 500 tokens). 17 MCP tools including semantic search (FTS5 + vector), code_context, code_impact, session bridges, crash recovery, and deployment safety gates. Saves 80-200K+ tokens per session by replacing file reads with targeted knowledge retrieval. Works with Claude Code and Codex CLI. Zero-config after install — hooks and MCP tools run automatically.

Fast semantic code search for AI agents — find symbols, references, and callers across any codebase. Pre-built index committed to git, instant queries via MCP

Security scanner MCP server that protects AI coding agents from generating vulnerable code. Uses tree-sitter AST analysis to detect OWASP Top 10 vulnerabilities, hardcoded secrets, and prompt injection attacks. Includes package hallucination detection across 4.3M packages (npm, PyPI, RubyGems, crates.io, pub.dev, CPAN, Raku). Works with Claude Desktop, Claude Code, Cursor, Cline, and any MCP-compatible client. Full feature list (for detailed descriptions): • 275+ security rules covering Python, JavaScript, TypeScript, Java, Go, Ruby, PHP, C/C++, Rust, C#, Terraform, Kubernetes • AST-based detection with tree-sitter (falls back to regex when unavailable) • Taint analysis for tracking user input to dangerous sinks • Package hallucination detection: verifies 4.3M+ packages across 7 ecosystems • Prompt injection detection: blocks exfiltration, jailbreaks, and malicious instructions • Automatic fix suggestions for common vulnerabilities • Zero config: works instantly with npx • CWE/OWASP metadata for compliance reporting

MCP Server that gives AI coding assistants instant access to your entire codebase through a persistent Tree-sitter index. Instead of grep flooding your context with 2000+ tokens per search, AiDex returns precise identifier matches in ~50 tokens. Supports 11 languages: C#, TypeScript, JavaScript, Rust, Python, C, C++, Java, Go, PHP, Ruby. Features include method signatures, cross-project search, time-based filtering, session notes, and an interactive browser-based viewer.

🔮 glyph – mcp server to map your codebase with symbol outlines

A library for indexing code repositories using Tree-sitter Stack Graphs, designed to be used in MCP servers and other tools.

MCP Server for Tree-sitter