Manticore Search MCP
Connect AI assistants to Manticore Search. Execute SQL queries, list tables, get schemas, and fetch documentation. Perfect for building RAG applications and search-powered AI agents.
README
Manticore Search MCP Server
MCP server for Manticore Search — enables AI assistants to query and manage Manticore Search databases directly.
Quick Start
Installation
# Option 1: Install with uv (recommended, requires PyPI release)
uvx mcp-manticore
# Option 2: Install with pip
pip install mcp-manticore
# Option 3: Run from source (for local development)
uvx --from . mcp-manticore
# Or: uv run mcp-manticore
Note:
uvxruns the package directly without installation. First-time run may take a moment to download dependencies.- The package must be published to PyPI for
uvx mcp-manticoreto work.- For local development or testing unreleased versions, use
uvx --from . mcp-manticore
What It Does
Tools
| Tool | Description |
|---|---|
run_query |
Execute SQL queries (SELECT, SHOW, DESCRIBE, etc.) |
list_tables |
List all tables and indexes |
describe_table |
Get table schema |
list_documentation |
List available documentation files |
get_documentation |
Fetch specific documentation from Manticore manual |
Prompts
manticore_initial_prompt— Built-in prompt teaching LLMs about Manticore Search features (full-text operators, KNN vector search, fuzzy search, etc.)
Health Check
HTTP endpoint at /health for monitoring connectivity.
Configuration
Environment Variables
| Variable | Default | Description |
|---|---|---|
MANTICORE_HOST |
localhost |
Manticore server host |
MANTICORE_PORT |
9308 |
HTTP API port |
MANTICORE_USER |
— | Username (optional) |
MANTICORE_PASSWORD |
— | Password (optional) |
MANTICORE_CONNECT_TIMEOUT |
30 |
Connection timeout (seconds) |
MANTICORE_QUERY_TIMEOUT |
30 |
Query timeout (seconds) |
MANTICORE_ALLOW_WRITE_ACCESS |
false |
Enable write operations (INSERT, UPDATE, DELETE) |
MANTICORE_ALLOW_DROP |
false |
Enable destructive operations (DROP, TRUNCATE) |
GITHUB_TOKEN |
— | GitHub token for higher API rate limit |
Safety
By default, all write operations are blocked. To enable:
# Enable writes (INSERT, UPDATE, DELETE)
export MANTICORE_ALLOW_WRITE_ACCESS=true
# Enable destructive operations (DROP, TRUNCATE)
export MANTICORE_ALLOW_DROP=true
Connect to Your AI Assistant
<details> <summary><strong>Claude Code</strong></summary>
Open terminal and run:
claude mcp add manticore -- uvx mcp-manticore
Or with environment variables:
claude mcp add manticore -- uvx mcp-manticore -- \
MANTICORE_HOST=localhost \
MANTICORE_PORT=9308
For full configuration, edit ~/.claude/mcp_settings.json:
{
"mcpServers": {
"manticore": {
"command": "uvx",
"args": ["mcp-manticore"],
"env": {
"MANTICORE_HOST": "localhost",
"MANTICORE_PORT": "9308"
}
}
}
}
Restart Claude Code or type /mcp restart to apply changes.
</details>
<details> <summary><strong>Cursor</strong></summary>
Method 1: Via Settings UI
- Open Cursor → Settings → Tools & MCP
- Click "Add MCP Server"
- Enter name:
manticore - Command:
uvx mcp-manticore
Method 2: Via Config File
Global config (~/.cursor/mcp.json):
{
"mcpServers": {
"manticore": {
"command": "uvx",
"args": ["mcp-manticore"],
"env": {
"MANTICORE_HOST": "localhost",
"MANTICORE_PORT": "9308"
}
}
}
}
Project config (.cursor/mcp.json in your project):
{
"mcpServers": {
"manticore": {
"command": "uvx",
"args": ["mcp-manticore"]
}
}
}
</details>
<details> <summary><strong>Windsurf</strong></summary>
Method 1: Via Cascade UI
- Open Windsurf → Cascade panel
- Click the MCPs icon (🔨) in the top-right
- Click "Add Server"
- Enter:
uvx mcp-manticore
Method 2: Via Config File
Edit ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"manticore": {
"command": "uvx",
"args": ["mcp-manticore"],
"env": {
"MANTICORE_HOST": "localhost",
"MANTICORE_PORT": "9308"
}
}
}
}
Or open directly in Windsurf: Cmd/Ctrl + Shift + P → "MCP Configuration Panel"
</details>
<details> <summary><strong>VS Code (with VSCode Copilot)</strong></summary>
VS Code uses the same MCP configuration as Cursor. Edit ~/.cursor/mcp.json:
{
"mcpServers": {
"manticore": {
"command": "uvx",
"args": ["mcp-manticore"],
"env": {
"MANTICORE_HOST": "localhost",
"MANTICORE_PORT": "9308"
}
}
}
}
</details>
<details> <summary><strong>Claude Desktop (Legacy)</strong></summary>
Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%/Claude/claude_desktop_config.json (Windows):
{
"mcpServers": {
"manticore": {
"command": "uvx",
"args": ["mcp-manticore"],
"env": {
"MANTICORE_HOST": "localhost",
"MANTICORE_PORT": "9308"
}
}
}
}
</details>
HTTP Transport (Remote MCP)
By default, MCP uses stdio (local). For remote access:
export MANTICORE_MCP_SERVER_TRANSPORT=http
export MANTICORE_MCP_BIND_PORT=8000
export MANTICORE_MCP_AUTH_TOKEN="your-secure-token"
uvx mcp-manticore
Connect via URL:
{
"mcpServers": {
"manticore": {
"url": "http://localhost:8000/mcp",
"headers": {
"Authorization": "Bearer your-secure-token"
}
}
}
}
Troubleshooting
Install uv (required)
macOS / Linux:
# Via installer script (recommended)
curl -LsSf https://astral.sh/uv/install.sh | sh
# Or via Homebrew
brew install uv
Windows:
# Via PowerShell
irm https://astral.sh/uv/install.ps1 | iex
# Or via winget
winget install astral-sh.uv
Verify installation:
uv --version
MCP server not connecting
- Verify Manticore is running:
curl http://localhost:9308/health - Check environment variables are set correctly
- For Claude Code: restart with
/mcp restart
Too many tools loaded
Some agents limit active MCP tools. Remove unused servers or use project-scoped configs.
Development
# Clone and setup
git clone https://github.com/manticoresoftware/mcp-manticore.git
cd mcp-manticore
# Install dependencies
uv sync
# Run locally
uv run mcp-manticore
# Run with custom config
MANTICORE_HOST=remote-server MANTICORE_PORT=9308 uv run mcp-manticore
# Run tests
uv run pytest
# Build package
uv build
# Publish to PyPI
uv publish
Architecture
| File | Purpose |
|---|---|
mcp_manticore/mcp_env.py |
Configuration management |
mcp_manticore/mcp_server.py |
MCP server implementation |
mcp_manticore/manticore_prompt.py |
LLM guidance/prompts |
mcp_manticore/docs_fetcher.py |
GitHub docs fetcher |
mcp_manticore/main.py |
CLI entry point |
License
<!-- Need to add this line for MCP registry publication --> <!-- mcp-name: io.github.manticoresoftware/mcp-manticore -->
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。