AutoDev Codebase MCP Server

@autodev/codebase

A platform-agnostic code analysis library with semantic search capabilities and MCP (Model Context Protocol) server support. This library provides intelligent code indexing, vector-based semantic search, and can be integrated into various development tools and IDEs.

🚀 Features

Semantic Code Search: Vector-based code search using embeddings
MCP Server Support: HTTP-based MCP server for IDE integration
Terminal UI: Interactive CLI with rich terminal interface
Tree-sitter Parsing: Advanced code parsing and analysis
Vector Storage: Qdrant vector database integration
Flexible Embedding: Support for various embedding models via Ollama

📦 Installation

1. Install and Start Ollama

# Install Ollama (macOS)
brew install ollama

# Start Ollama service
ollama serve

# In a new terminal, pull the embedding model
ollama pull dengcao/Qwen3-Embedding-0.6B:Q8_0

2. Install ripgrep

ripgrep is required for fast codebase indexing. Install it with:

# Install ripgrep (macOS)
brew install ripgrep

# Or on Ubuntu/Debian
sudo apt-get install ripgrep

# Or on Arch Linux
sudo pacman -S ripgrep

3. Install and Start Qdrant

Start Qdrant using Docker:

# Start Qdrant container
docker run -p 6333:6333 -p 6334:6334 qdrant/qdrant

Or download and run Qdrant directly:

# Download and run Qdrant
wget https://github.com/qdrant/qdrant/releases/latest/download/qdrant-x86_64-unknown-linux-gnu.tar.gz
tar -xzf qdrant-x86_64-unknown-linux-gnu.tar.gz
./qdrant

4. Verify Services Are Running

# Check Ollama
curl http://localhost:11434/api/tags

# Check Qdrant
curl http://localhost:6333/collections

5. Install Autodev-codebase

npm install -g @autodev/codebase

Alternatively, you can install it locally:

git clone https://github.com/anrgct/autodev-codebase
cd autodev-codebase
npm install
npm run build
npm link

🛠️ Usage

Command Line Interface

The CLI provides two main modes:

1. Interactive TUI Mode (Default)

# Basic usage: index your current folder as the codebase.
# Be cautious when running this command if you have a large number of files.
codebase


# With custom options
codebase --demo # Create a local demo directory and test the indexing service, recommend for setup
codebase --path=/my/project
codebase --path=/my/project --log-level=info

2. MCP Server Mode (Recommended for IDE Integration)

# Start long-running MCP server
cd /my/project
codebase mcp-server

# With custom configuration
codebase mcp-server --port=3001 --host=localhost
codebase mcp-server --path=/workspace --port=3002

⚙️ Configuration

Configuration Files & Priority

The library uses a layered configuration system, allowing you to customize settings at different levels. The priority order (highest to lowest) is:

CLI Parameters (e.g., --model, --ollama-url, --qdrant-url, --config, etc.)
Project Config File (./autodev-config.json)
Global Config File (~/.autodev-cache/autodev-config.json)
Built-in Defaults

Settings specified at a higher level override those at lower levels. This lets you tailor the behavior for your environment or project as needed.

Config file locations:

Global: ~/.autodev-cache/autodev-config.json
Project: ./autodev-config.json
CLI: Pass parameters directly when running commands

Global Configuration

Create a global configuration file at ~/.autodev-cache/autodev-config.json:

{
  "isEnabled": true,
  "embedder": {
    "provider": "ollama",
    "model": "dengcao/Qwen3-Embedding-0.6B:Q8_0",
    "dimension": 1024,
    "baseUrl": "http://localhost:11434"
  },
  "qdrantUrl": "http://localhost:6333",
  "qdrantApiKey": "your-api-key-if-needed",
  "searchMinScore": 0.4
}

Project Configuration

Create a project-specific configuration file at ./autodev-config.json:

{
  "embedder": {
    "provider": "openai-compatible",
    "apiKey": "sk-xxxxx",
    "baseUrl": "http://localhost:2302/v1",
    "model": "openai/text-embedding-3-smallnpm",
    "dimension": 1536,
  },
  "qdrantUrl": "http://localhost:6334"
}

Configuration Options

Option	Type	Description	Default
`isEnabled`	boolean	Enable/disable code indexing feature	`true`
`embedder.provider`	string	Embedding provider (`ollama`, `openai`, `openai-compatible`)	`ollama`
`embedder.model`	string	Embedding model name	`dengcao/Qwen3-Embedding-0.6B:Q8_0`
`embedder.dimension`	number	Vector dimension size	`1024`
`embedder.baseUrl`	string	Provider API base URL	`http://localhost:11434`
`embedder.apiKey`	string	API key (for OpenAI/compatible providers)	-
`qdrantUrl`	string	Qdrant vector database URL	`http://localhost:6333`
`qdrantApiKey`	string	Qdrant API key (if authentication enabled)	-
`searchMinScore`	number	Minimum similarity score for search results	`0.4`

Note: The isConfigured field is automatically calculated based on the completeness of your configuration and should not be set manually. The system will determine if the configuration is valid based on the required fields for your chosen provider.

Configuration Priority Examples

# Use global config defaults
codebase

# Override model via CLI (highest priority)
codebase --model="custom-model"

# Use project config with CLI overrides
codebase --config=./my-config.json --qdrant-url=http://remote:6333

🔧 CLI Options

Global Options

--path=<path> - Workspace path (default: current directory)
--demo - Create demo files in workspace
--force - ignore cache force re-index
--ollama-url=<url> - Ollama API URL (default: http://localhost:11434)
--qdrant-url=<url> - Qdrant vector DB URL (default: http://localhost:6333)
--model=<model> - Embedding model (default: nomic-embed-text)
--config=<path> - Config file path
--storage=<path> - Storage directory path
--cache=<path> - Cache directory path
--log-level=<level> - Log level: error|warn|info|debug (default: error)
--log-level=<level> - Log level: error|warn|info|debug (default: error)
--help, -h - Show help

MCP Server Options

--port=<port> - HTTP server port (default: 3001)
--host=<host> - HTTP server host (default: localhost)

IDE Integration (Cursor/Claude)

Configure your IDE to connect to the MCP server:

{
  "mcpServers": {
    "codebase": {
      "url": "http://localhost:3001/sse"
    }
  }
}

For clients that do not support SSE MCP, you can use the following configuration:

{
  "mcpServers": {
    "codebase": {
      "command": "codebase",
      "args": [
        "stdio-adapter",
        "--server-url=http://localhost:3001/sse"
      ]
    }
  }
}

🌐 MCP Server Features

Web Interface

Home Page: http://localhost:3001 - Server status and configuration
Health Check: http://localhost:3001/health - JSON status endpoint
MCP Endpoint: http://localhost:3001/sse - SSE/HTTP MCP protocol endpoint

Available MCP Tools

search_codebase - Semantic search through your codebase
- Parameters: query (string), limit (number), filters (object)
- Returns: Formatted search results with file paths, scores, and code blocks

Scripts

# Development mode with demo files
npm run dev

# Build for production
npm run build

# Type checking
npm run type-check

# Run TUI demo
npm run demo-tui

# Start MCP server demo
npm run mcp-server

Embedding Models PK

Mainstream Embedding Models Performance

Model	Dimension	Avg Precision@3	Avg Precision@5	Good Queries (≥66.7%)	Failed Queries (0%)
siliconflow/Qwen/Qwen3-Embedding-8B	4096	76.7%	66.0%	5/10	0/10
siliconflow/Qwen/Qwen3-Embedding-4B	2560	73.3%	54.0%	5/10	1/10
voyage/voyage-code-3	1024	73.3%	52.0%	6/10	1/10
siliconflow/Qwen/Qwen3-Embedding-0.6B	1024	63.3%	42.0%	4/10	1/10
morph-embedding-v2	1536	56.7%	44.0%	3/10	1/10
openai/text-embedding-ada-002	1536	53.3%	38.0%	2/10	1/10
voyage/voyage-3-large	1024	53.3%	42.0%	3/10	2/10
openai/text-embedding-3-large	3072	46.7%	38.0%	1/10	3/10
voyage/voyage-3.5	1024	43.3%	38.0%	1/10	2/10
voyage/voyage-3.5-lite	1024	36.7%	28.0%	1/10	2/10
openai/text-embedding-3-small	1536	33.3%	28.0%	1/10	4/10
siliconflow/BAAI/bge-large-en-v1.5	1024	30.0%	28.0%	0/10	3/10
siliconflow/Pro/BAAI/bge-m3	1024	26.7%	24.0%	0/10	2/10
ollama/nomic-embed-text	768	16.7%	18.0%	0/10	6/10
siliconflow/netease-youdao/bce-embedding-base_v1	1024	13.3%	16.0%	0/10	6/10

Ollama-based Embedding Models Performance

Model	Dimension	Precision@3	Precision@5	Good Queries (≥66.7%)	Failed Queries (0%)
ollama/dengcao/Qwen3-Embedding-4B	2560	66.7%	48.0%	4/10	1/10
ollama/dengcao/Qwen3-Embedding-0.6B	1024	63.3%	44.0%	3/10	0/10
ollama/dengcao/Qwen3-Embedding-0.6B	1024	63.3%	44.0%	3/10	0/10
ollama/dengcao/Qwen3-Embedding-4B	2560	60.0%	48.0%	3/10	1/10
lmstudio/taylor-jones/bge-code-v1-Q8_0-GGUF	1536	60.0%	54.0%	4/10	1/10
ollama/dengcao/Qwen3-Embedding-8B	4096	56.7%	42.0%	2/10	2/10
ollama/hf.co/nomic-ai/nomic-embed-code-GGUF	3584	53.3%	44.0%	2/10	0/10
ollama/bge-m3	1024	26.7%	24.0%	0/10	2/10
ollama/hf.co/nomic-ai/nomic-embed-text-v2-moe-GGUF	768	26.7%	20.0%	0/10	2/10
ollama/granite-embedding:278m-fp16	768	23.3%	18.0%	0/10	4/10
ollama/unclemusclez/jina-embeddings-v2-base-code	768	23.3%	16.0%	0/10	5/10
lmstudio/awhiteside/CodeRankEmbed-Q8_0-GGUF	768	23.3%	16.0%	0/10	5/10
lmstudio/wsxiaoys/jina-embeddings-v2-base-code-Q8_0-GGUF	768	23.3%	16.0%	0/10	5/10
ollama/dengcao/Dmeta-embedding-zh	768	20.0%	20.0%	0/10	6/10
ollama/znbang/bge.5-q8_0	384	16.7%	16.0%	0/10	6/10
lmstudio/nomic-ai/nomic-embed-text-v1.5-GGUF@Q4_K_M	768	16.7%	14.0%	0/10	6/10
ollama/nomic-embed-text	768	16.7%	18.0%	0/10	6/10
ollama/snowflake-arctic-embed2:568m	1024	16.7%	18.0%	0/10	5/10

This server cannot be installed

security - not tested

license - not found

quality - not tested

How are these scores calculated?

remote-capable server

The server can be hosted and run remotely because it primarily relies on remote services or has no dependency on the local environment.

HTTP-based server that provides semantic code search capabilities to IDEs through the Model Context Protocol, allowing efficient codebase exploration without repeated indexing.

Related MCP Servers

MCP Code Indexer
zxfgds
-
security
F
license
-
quality
A smart code retrieval tool based on Model Context Protocol that provides efficient and accurate code repository search capabilities for large language models.
Last updated -
Python
code-index-mcp
johnhuang316
A
security
A
license
A
quality
A Model Context Protocol (MCP) server that helps large language models index, search, and analyze code repositories with minimal setup
Last updated -
11
93
Python
MIT License
Better Qdrant MCP Server
wrediam
-
security
A
license
-
quality
A Model Context Protocol server that enables semantic search capabilities by providing tools to manage Qdrant vector database collections, process and embed documents using various embedding services, and perform semantic searches across vector embeddings.
Last updated -
89
TypeScript
MIT License
Docs MCP Server
buger
A
security
A
license
A
quality
A flexible Model Context Protocol server that makes documentation or codebases searchable by AI assistants, allowing users to chat with code or docs by simply pointing to a git repository or folder.
Last updated -
1
22
36
JavaScript
MIT License

View all related MCP servers

AutoDev Codebase MCP Server