MCP-Manticore: AI for queries to Manticore Search

MCP-Manticore: AI Assistants Meet Manticore Search via Model Context Protocol

MCP-Manticore is a Model Context Protocol server that grants direct access to Manticore Search for AI assistants. This tool enables Cursor, Claude Code, Codex CLI, and other MCP-compatible clients to analyze data schemas, review documentation, and execute SQL queries in real time—eliminating syntax hallucinations and query errors in full-text, vector, or fuzzy search scenarios.

AI receives context before generating code: it reads table structures, studies available functions, and tests queries against a live server. Support for stdio, HTTP, SSE, and JWT authentication makes integration flexible and production-ready.

Use Cases

Application Development

AI automates repetitive tasks:

• Creating tables tailored to Manticore’s requirements.
• Building complex queries with JOINs, aggregations, and full-text search.
• Debugging functions like MATCH AGAINST or vector embeddings.

Data Analysis

Direct AI-driven queries simplify exploration:

• "List all tables" → calls list_tables().
• "Describe column structure" → schema introspection.
• "Find similar records" → generates queries using PERCENT_RANK or hybrid search.

Without MCP: documentation → query → error → fix. With MCP: question → schema check → execution → result.

Technical Implementation

MCP-Manticore provides tools for:

• Reading Manticore documentation.
• Schema introspection (tables, columns, indexes).
• Executing SELECT, INSERT (optional), with DROP protection.

Supported clients:

• Cursor.
• Claude Code.
• Codex CLI.
• Windsurf.
• Any MCP-compliant client.

Transport: stdio for CLI, HTTP/SSE for web interfaces.

Installation & Configuration

Prerequisites

Install uv for rapid startup:

curl -LsSf https://astral.sh/uv/install.sh | sh

uvx automatically downloads and runs mcp-manticore.

Environment Variables

| Variable | Description | Default |

|----------|-------------|---------|

| MANTICORE_HOST | Server host | localhost |

| MANTICORE_PORT | HTTP port | 9308 |

| MANTICORE_ALLOW_WRITE_ACCESS | Write access permission | false |

| MANTICORE_ALLOW_DROP | DROP permission | false |

| MANTICORE_MCP_TRANSPORT | Transport method | stdio |

| MANTICORE_MCP_AUTH_TOKEN | JWT token | — |

Example:

export MANTICORE_HOST=localhost
export MANTICORE_PORT=9308
export MANTICORE_ALLOW_WRITE_ACCESS=true
export MANTICORE_ALLOW_DROP=false

MCP Client Configuration

Add to your JSON config:

{
  "mcpServers": {
    "manticore": {
      "command": "uvx",
      "args": ["mcp-manticore"],
      "env": {
        "MANTICORE_HOST": "localhost",
        "MANTICORE_PORT": "9308"
      }
    }
  }
}

Testing

Ask AI: "Show me all tables in Manticore." The list_tables() function returns the result.

Future Outlook

Autonomous AI agents will soon be able to:

• Scan GitHub repositories.
• Auto-install MCP-Manticore.
• Connect to databases independently.

This shifts the paradigm from manual SQL to declarative data interaction.

Key Takeaways

• Direct schema access: AI avoids Manticore syntax errors.
• Flexible security: Separate controls for WRITE and DROP operations.
• Multi-transport support: stdio, HTTP, SSE for diverse clients.
• Open standard: Full compatibility with the broader MCP ecosystem.
• Automation boost: Accelerates dev and data exploration by 50–70%.

— Editorial Team