CodeGB
Provides a local code knowledge graph for Java projects, enabling querying of classes, methods, fields, calls, inheritance, and imports via MCP tools like query, context, impact, and cypher.
README
CodeGB
⚠️ This repository has moved
CodeGB is no longer maintained.
The project has been renamed and continued as Tally.
➡️ New repository: https://github.com/Killian-Filippov/Tally
This repository is archived and kept only for historical reference.
CodeGB is a code knowledge graph tool for local developer environments. It provides an MCP Server that runs on your machine and can be called by MCP Clients (such as Claude/Cursor).
Feature Overview
- Java code parsing and graph construction (classes, methods, fields, calls, inheritance, imports)
- Tree-sitter-first Java extraction, with regex kept only as a compatibility fallback
- MCP tools:
querycontextimpactcypherlist_repos
- Local-first storage and querying
- Switchable graph storage backend (WASM / Native)
Requirements
- Node.js 18+
- pnpm
Install Dependencies
pnpm install
Build
pnpm build
Local Usage (CLI)
# 1) Initialize the repository index directory
pnpm exec tsx packages/cli/src/index.ts init /path/to/your/repo --storage .javakg
# 2) Build index
pnpm exec tsx packages/cli/src/index.ts index /path/to/your/repo --storage .javakg
# 2.1) Incremental indexing (only process changed Java files in git diff)
pnpm exec tsx packages/cli/src/index.ts index /path/to/your/repo --storage .javakg --changed-files
# 3) Query
pnpm exec tsx packages/cli/src/index.ts query "payment" --storage .javakg --limit 10
Start MCP Server (stdio)
JAVA_KG_DB_PATH=.javakg pnpm exec tsx packages/mcp-server/src/cli.ts
Notes:
- This process communicates with the MCP Client via stdio.
- In your MCP Client, configure this command as the MCP Server startup command.
MCP Client Config Template (Minimal Working Setup)
Use the following template consistently:
command:pnpmargs:["exec", "tsx", "packages/mcp-server/src/cli.ts"]env.JAVA_KG_DB_PATH: your index directory (must match--storage)env.CODEGB_AUTO_INDEX_INTERVAL_MS: polling interval (ms) for MCP background incremental indexing (default3000, set to0to disable)env.CODEGB_MCP_CACHE_TTL_MS: cache TTL (ms) for MCPquery/context(default60000, set to0to disable)env.CODEGB_MCP_CACHE_L1_MAX_ENTRIES: in-process L1 cache size (default256, set to0to disable)env.CODEGB_MCP_CACHE_L2_MAX_ENTRIES: persistent L2 cache size (default4096, set to0to disable)
Replace "/ABS/PATH/TO/CodeGB" and "/ABS/PATH/TO/.javakg" with absolute paths.
Claude Desktop
The config file (macOS) is usually at ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"codegb": {
"command": "pnpm",
"args": ["exec", "tsx", "packages/mcp-server/src/cli.ts"],
"cwd": "/ABS/PATH/TO/CodeGB",
"env": {
"JAVA_KG_DB_PATH": "/ABS/PATH/TO/.javakg"
}
}
}
}
Cursor
Workspace file .cursor/mcp.json:
{
"mcpServers": {
"codegb": {
"command": "pnpm",
"args": ["exec", "tsx", "packages/mcp-server/src/cli.ts"],
"cwd": "/ABS/PATH/TO/CodeGB",
"env": {
"JAVA_KG_DB_PATH": "/ABS/PATH/TO/.javakg"
}
}
}
}
VS Code
Workspace file .vscode/mcp.json:
{
"servers": {
"codegb": {
"type": "stdio",
"command": "pnpm",
"args": ["exec", "tsx", "packages/mcp-server/src/cli.ts"],
"cwd": "/ABS/PATH/TO/CodeGB",
"env": {
"JAVA_KG_DB_PATH": "/ABS/PATH/TO/.javakg"
}
}
}
}
First-Time Indexing (Required Before MCP)
MCP only provides queries. For first-time usage, run init + index first:
pnpm exec tsx packages/cli/src/index.ts init /ABS/PATH/TO/REPO --storage /ABS/PATH/TO/.javakg
pnpm exec tsx packages/cli/src/index.ts index /ABS/PATH/TO/REPO --storage /ABS/PATH/TO/.javakg
Java Parser Runtime
- The indexing pipeline uses
tree-sitteras the primary Java parser and extraction path. JAVA_QUERIESis the authoritative query template used by the tree-sitter extractor for symbol and relation capture.- The legacy regex extractor is still present only as a compatibility fallback when the tree-sitter runtime is unavailable or a single-file tree-sitter parse fails during indexing.
- Operationally, you should treat CodeGB as a tree-sitter-based Java indexer, not as a dual-parser system.
FAQ
- Q: MCP connects successfully, but returns empty results?
A: Usually initial indexing was not completed, or
JAVA_KG_DB_PATHdoes not match index--storage. - Q: Startup fails with JSON error codes?
A:
E_NODE_VERSION(Node < 18),E_STORAGE_PERM(directory not writable),E_WORKER_UNAVAILABLE(worker unavailable),E_BACKEND_INIT(backend initialization failed). - Q:
pnpmnot found? A: Ensurepnpmis available in the client runtimePATH, or use an absolute path topnpm. - Q:
list_reposshows old repositories? A: Re-runinit + indexfor the new repository and updateJAVA_KG_DB_PATH.
Fallback Behavior
- Backend fallback: when
CODEGB_DB_BACKEND=auto,wasmis preferred, and if it fails it automatically falls back tonative, with a one-time diagnostic log. - Cypher fallback: the
cyphertool prefers backend execution; when backend is unavailable or returns incompatible results, it automatically falls back to the in-memory graph compatible execution path. - Data fallback: if the index directory is empty or graph loading fails, the service starts with an empty graph and tools return empty results instead of crashing the process.
Graph Database Backend Selection
The current implementation supports two paths:
kuzu-wasm: better cross-platform compatibility and installation experience, suitable for default distribution.kuzu(native): may be faster on some machines, but has higher installation/compatibility cost.
Unified switch:
CODEGB_DB_BACKEND=wasm|native|auto- Default:
wasm(compatibility first) auto: trywasmfirst, automatically fall back tonativeon failure, and print a one-time diagnostic log.
Note: backend switching has been implemented in the core layer. The concrete runtime strategy can be managed uniformly through startup configuration.
Tests
# Core tests
pnpm exec tsx --test packages/core/test/*.test.ts
# E2E tests
pnpm test:e2e
Release Gate
Before publishing a developer-trial version, run:
pnpm run release:gate
Gate conditions:
- Core e2e passes (
pnpm test:e2e:phase1) benchmark.mdexistsdocs/release-notes.mdcontains the current version heading (## v<version>)
Project Structure
packages/core: parsing, graph models, storage adapters, search, MCP tool logicpackages/mcp-server: MCP Server startup and protocol integrationpackages/cli: local init/index/query commandstests/e2e: end-to-end tests
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.