Open-source Rust knowledge graph for AI coding agents.

Fast indexing. Deep code understanding. Embedded web UI. Single binary.

codryn is an open-source Rust knowledge graph and MCP server for AI coding agents, built to make large codebases easier to explore, trace, and understand.

If this project is useful, give it a star. It helps more people discover the project and support continued open-source work.

Install

curl -fsSL https://raw.githubusercontent.com/WolfCanCode/Codryn/main/install.sh | sh

Tries pre-built binaries first. Falls back to building from source if none are available for your platform. Requires Rust and Node.js 20+ for source builds.

Why this project

AI coding agents are much better when they can understand structure, not just grep files.

codryn indexes a repository into a persistent graph of:

• functions, classes, methods, files, and folders
• call paths and imports
• routes, DTOs, and service layers
• frontend component relationships and dependency injection
• CI/CD pipelines, jobs, stages, includes, and dependency edges
• infrastructure resources from Docker, Kubernetes, Kustomize, Helm, and Terraform-style manifests
• cross-project links between systems

That gives agents fast answers for things like:

• where a request starts and where it ends
• who calls this function
• what breaks if I change this symbol
• which files matter next
• how frontend and backend connect

Highlights

• Open source + MIT licensed
• Rust implementation with a strong focus on speed and portability
• Persistent knowledge graph stored in SQLite
• Embedded dashboard for visual graph exploration
• Incremental indexing so re-runs stay fast
• Cross-project linking for multi-repo systems
• 46 MCP tools for search, tracing, navigation, analysis, architecture, and agent-first operations
• Semantic search using all-MiniLM-L6-v2 embeddings for natural language code queries
• CI/CD and infrastructure discovery for GitHub Actions, GitLab CI, CircleCI, Azure Pipelines, Bitbucket Pipelines, Jenkinsfile-style jobs, Docker, Kubernetes, Kustomize, Helm, and Terraform resources
• 64 language detection plus tree-sitter walkers for Rust, TypeScript, JavaScript, Python, Go, Java, Kotlin, Dart, Lua, Haskell, C/C++, C#, Ruby, PHP, Swift, Scala, Elixir, and Bash
• Agent-first tools: what_if, ask_graph, plan_refactoring, detect_patterns, semantic_search, generate_openapi
• Confidence scoring on every graph edge (0.0–1.0 provenance)
• Single binary with no Docker required

What makes it interesting

• Built for AI agents: not just search, but graph-aware navigation and analysis
• Framework-aware: strong support for Spring Boot, Angular, Vue, FastAPI, Go, and Next.js
• Useful locally: run as an MCP server or open the web UI
• Practical architecture: one workspace, one binary, one local graph store

MCP tools

Some of the most useful tools (46 total):

Full tool reference: docs/MCP_TOOLS.md

CLI

codryn                        Run as MCP server on stdin/stdout
codryn install                Auto-configure coding agents (interactive)
codryn status                 Show agent installation status
codryn activate [--global]    Activate steering for workspace
codryn query <tool> [--args]  Run MCP tool as CLI command
codryn validate --project P   Validate graph consistency
codryn dedupe --project P     Deduplicate graph nodes
codryn complexity --project P Report complex symbols
codryn deps --project P       List/check dependencies
codryn backup [path]          Back up the graph database
codryn --ui [--port=N]        Enable web UI (default port 9749)

Architecture

crates/
├── codryn-foundation    — Config, platform detection, utilities
├── codryn-store         — SQLite graph store with pooling & embeddings
├── codryn-discover      — 64-language file discovery
├── codryn-pipeline      — Multi-pass indexing with checkpoints
├── codryn-treesitter    — 30+ AST walkers
├── codryn-graph-buffer  — Batch graph operations with confidence
├── codryn-cypher        — Cypher → SQL query engine
├── codryn-services      — Navigation, flow, analysis, semantic search
├── codryn-mcp           — 46 MCP tool handlers
├── codryn-cli           — Interactive install, validate, query, etc.
├── codryn-ui            — Web dashboard (React + embedded assets)
├── codryn-watcher       — File system watcher
├── codryn-bench         — Performance benchmarks
└── codryn-bin           — App binary

ui/
└── React/Vite dashboard for graph exploration

Local run

cargo build --release
./target/release/codryn --ui

Then open http://127.0.0.1:9749.

Supported languages

Supports 64 language mappings, including Rust, TypeScript, JavaScript, Java, Kotlin, Python, Go, Dart, Lua, Haskell, C, C++, C#, PHP, Ruby, Scala, Swift, Elixir, SQL, HTML, CSS, Vue, Svelte, Bash, Dockerfile, YAML, and more. Deep tree-sitter walkers for 30+ languages with error-tolerant AST recovery.

Open source

This project is MIT licensed and intended to be easy to use, inspect, extend, and contribute to.

License

MIT

References

• Research paper: persistent knowledge graphs for AI coding agents