OpenTrace

A knowledge graph that maps your codebase structure, service architecture, and system relationships — then exposes it all through MCP so AI tools can understand your systems.

Index any repo, any size, in seconds. Build a knowledge graph and watch your LLM fetch true codebase context in realtime.

What It Does

OpenTrace indexes source code and builds a queryable knowledge graph. Point it at a repo and it will:

1. Parse every file using tree-sitter WASM grammars (12 languages)
2. Extract classes, functions, imports, and call relationships
3. Build a knowledge graph stored in LadybugDB (embedded graph database)
4. Summarize every node using template-based identifier analysis
5. Expose the graph via MCP tools to any compatible AI client

Get Started

Linux/MacOS

uvx opentraceai index .   # Index a local project for use with MCP etc

Claude Plugin

/plugin marketplace add https://github.com/opentrace/opentrace
/plugin install opentrace-oss@opentrace-oss
/reload-plugins

More info: https://opentrace.github.io/opentrace/getting-started/install-plugin/

OpenCode Plugin

Add to ~/.config/opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["@opentrace/opencode"]
}

More info: https://opentrace.github.io/opentrace/getting-started/install-opencode/

Gemini CLI

gemini mcp add opentraceai uvx opentraceai mcp

Run Completely in the Browser

No install: app.opentrace.ai

Run UI Locally

git clone https://github.com/opentrace/opentrace.git && cd opentrace
make install && make ui  # Runs on http://localhost:5173/

To enable knowledge vaults (LLM-compiled markdown wikis), also run the
agent's REST server in a second terminal and open the UI with a ?server=
query param so it reads from the backend:

uv --project agent run opentraceai serve   # listens on http://localhost:8787
# then open: http://localhost:5173/?server=http://localhost:8787

See Wiki & Vaults for the compile flow and Wiki Providers for the supported LLM providers (Anthropic, Gemini, OpenAI, local).

More Information

Full documentation: opentrace.github.io/opentrace — install guides, architecture, and reference.

Prerequisites: plugins and CLI need uv; OpenCode plugin runs in Bun (provided by OpenCode itself). Source build also needs Node 22+ and Python 3.12+. See Troubleshooting if anything fails.

Claude Code Plugin

The plugin gives Claude Code 5 agents, 4 slash commands, and MCP graph tools.

Full details: Claude Code Plugin reference.

OpenCode Plugin

The plugin gives OpenCode 9 native tools and three hooks (system-prompt, tool-execute, auth) — calls the opentraceai CLI directly rather than going through MCP.

Full details: OpenCode Plugin reference.

Supported Languages

Full extraction (symbols + calls + imports): Python, TypeScript/JavaScript, Go
Structural extraction (symbols only): Rust, Java, Kotlin, C#, C/C++, Ruby, Swift
Indexed as file nodes: JSON, YAML, TOML, Protobuf, SQL, GraphQL, Bash

Full language matrix: Supported Languages.

Architecture

┌───────────────────────────────────────────────────────────┐
│                      UI (React/TS)                        │
│            Browser-based indexer + graph explorer         │
│               localhost:5173 / app.opentrace.ai           │
│                                                           │
│  ┌───────────────┐  ┌─────────────────┐  ┌──────────────┐ │
│  │  Web Worker   │  │  LadybugDB WASM │  │  Chat Agent  │ │
│  │  tree-sitter  │  │  (embedded      │  │  LLM-powered │ │
│  │  WASM parsers │  │   graph store)  │  │  graph tools │ │
│  └───────────────┘  └─────────────────┘  └──────────────┘ │
└───────────────────────────────────────────────────────────┘

┌───────────────────────────────────────────────────────────┐
│                 Claude Code Plugin (MCP)                  │
│     Python agent exposes graph tools via MCP protocol     │
│         uvx opentraceai mcp  →  stdio transport           │
└───────────────────────────────────────────────────────────┘

┌───────────────────────────────────────────────────────────┐
│                  OpenCode Plugin (native)                 │
│      TS plugin in OpenCode's Bun runtime — 9 tools +      │
│         system-prompt, tool-execute, auth hooks           │
│           opentraceai CLI  →  spawn (no MCP)              │
└───────────────────────────────────────────────────────────┘

See Architecture Overview for details.

Repository Structure

ui/                   — React/TypeScript frontend (graph explorer + browser indexer)
agent/                — Python agent (CLI + MCP server for Claude Code)
proto/                — Protobuf definitions (shared schema)
plugins/              — Editor / AI integrations
  claude-code/        — Claude Code plugin (agents, commands, skills, hooks)
  opencode/           — OpenCode plugin (native TS, Bun runtime)
docs/                 — Documentation site (mkdocs-material)
tests/                — Cross-validation test fixtures
benchmark/            — Accuracy benchmarks

Development

See Development Setup for prerequisites, commands, and the full dev workflow.

Contributing

See CONTRIBUTING.md for guidelines on how to contribute.

License

Apache License 2.0 — see LICENSE for details.

{
  "$schema": "https://anthropic.com/claude-code/marketplace.schema.json",
  "name": "opentrace-oss",
  "description": "Knowledge graph tools for exploring system architecture, code structure, and service dependencies",
  "owner": {
    "name": "OpenTrace"
  },
  "plugins": [
    {
      "name": "opentrace-oss",
      "description": "OpenTrace knowledge graph — search, traverse, and analyze indexed codebases",
      "version": "0.7.0",
      "author": {
        "name": "OpenTrace"
      },
      "source": "./plugins/claude-code",
      "category": "development"
    }
  ]
}