SochDB MCP Server

Minimal Model Context Protocol (MCP) server for SochDB - an AI-native database.

Overview

This is a thin adapter layer that exposes SochDB's AI-native features via the MCP protocol, allowing LLM clients like Claude Desktop, Cursor, and ChatGPT to interact with your database.

MCP Client (Claude, Cursor, etc.)
     │
     │ JSON-RPC over stdio
     ▼
┌─────────────────────────────────┐
│  sochdb-mcp                     │
│  - Stdio framing (~50 lines)    │
│  - JSON-RPC dispatch            │
│  - MCP methods                  │
└─────────────────────────────────┘
     │
     │ Direct Rust calls
     ▼
┌─────────────────────────────────┐
│  SochDB                         │
│  - Context queries → AI context │
│  - TOON format → token savings  │
│  - Path-based access            │
└─────────────────────────────────┘

Installation

# Build from source
cargo build --release --package sochdb-mcp

# Binary is at target/release/sochdb-mcp

Usage

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "sochdb": {
      "command": "/path/to/sochdb-mcp",
      "args": ["--db", "/path/to/your/database"]
    }
  }
}

Cursor

Add to your .cursor/mcp.json:

{
  "mcpServers": {
    "sochdb": {
      "command": "/path/to/sochdb-mcp",
      "args": ["--db", "./data"]
    }
  }
}

Command Line Testing

# Test the server manually
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}' | sochdb-mcp --db ./test_data

Available Tools

sochdb.context_query

Fetch AI-optimized context with token budgeting. This is SochDB's killer feature for LLMs.

{
  "name": "sochdb.context_query",
  "arguments": {
    "sections": [
      {
        "name": "recent_messages",
        "priority": 10,
        "kind": "last",
        "table": "messages",
        "top_k": 5
      },
      {
        "name": "user_profile",
        "priority": 20,
        "kind": "get",
        "path": "users/current"
      }
    ],
    "token_budget": 2000
  }
}

sochdb.query

Execute SochQL queries directly.

{
  "name": "sochdb.query",
  "arguments": {
    "query": "SELECT * FROM users WHERE active = true"
  }
}

sochdb.get / sochdb.put / sochdb.delete

Path-based CRUD operations with O(|path|) complexity.

{
  "name": "sochdb.get",
  "arguments": {
    "path": "users/123/profile/name"
  }
}

sochdb.list_tables / sochdb.describe

Schema introspection tools.

Architecture

The server is intentionally minimal:

• ~100 lines for stdio framing
• ~100 lines for JSON-RPC types
• ~200 lines for MCP protocol handling
• ~200 lines for tool execution

No external MCP framework. Just serde_json + SochDB.

Protocol Support

• Protocol version: 2024-11-05
• Transport: stdio (stdin/stdout with Content-Length framing)
• Capabilities:
  • tools: list and call
  • resources: list and read (exposes tables)

Why This Approach?

1. SochDB is already AI-native: Operations, SochQL, context engine, and token budgets are built-in. The MCP layer just exposes them.

2. Thin is good: The MCP server doesn't need to understand schemas or indexing. It translates between JSON-RPC and SochDB calls.

3. No framework bloat: A few hundred lines of code vs. pulling in a heavy MCP SDK.

Logging

All logs go to stderr (required for stdio transport - stdout is for protocol messages only).

Set RUST_LOG=sochdb_mcp=debug for verbose output.

License

Apache-2.0