Epic: MCP Server for Data Studio ecosystem (dockit + sqlkit) #9

Epic: MCP Server for Data Studio ecosystem (dockit + sqlkit)

Status: Planning
Target release: TBD

Background

data-studio-agent powers AI agent loops for both dockit (NoSQL: Elasticsearch, OpenSearch, MongoDB, DynamoDB) and sqlkit (SQL: PostgreSQL, MySQL, SQL Server, SQLite). Currently, only the built-in Tauri-based AI assistant can use these capabilities.

The Model Context Protocol (MCP) has become the standard for AI coding agents (Claude Code, Cursor, Windsurf, OpenCode, Codex) to interact with external tools. This epic delivers a unified MCP Server that exposes both dockit and sqlkit capabilities to any MCP-compatible agent.

Architecture

code agent (Claude Code / Cursor / OpenCode ...)
    |
    | MCP stdio protocol
    v
@geekfun/data-studio-mcp (npm package, pure TypeScript)
    |
    | @modelcontextprotocol/sdk
    | node:http
    |
    |  POST /invoke    POST /tools
    +------+----------------+---
           |                |
           v                v
       dockit:9120    sqlkit:9121
       (HTTP server   (HTTP server
        in dockit)     in sqlkit)
           |                |
           v                v
       invoke_capability_inner()
       + get_available_tools()
           |                |
           v                v
       dockit capabilities     sqlkit capabilities
       (es__*, mongo__*,       (sql__execute_query,
        dynamo__*, etc.)        sql__list_tables, etc.)

Two layers only:

• MCP Server: pure TypeScript, depends only on @modelcontextprotocol/sdk + built-in node:http
• Bridge: dockit and sqlkit each expose a lightweight HTTP server that delegates to their existing invoke_capability_inner()

No Rust binary, no platform-specific builds, no launcher shim. The MCP server is just an npm package that npx runs directly.

Repository structure changes

data-studio-agent/
+-- Cargo.toml                  [workspace] (existing, single member)
+-- package.json                npm workspace root (new)
|
+-- crates/
|   +-- data-studio-agent/      existing Rust code, moved from root
|       +-- Cargo.toml          name = "data-studio-agent" (unchanged)
|       +-- src/                existing code, no changes needed
|
+-- packages/
|   +-- data-studio-mcp/        NEW: MCP Server (pure TypeScript)
|       +-- package.json        @geekfun/data-studio-mcp
|       +-- tsconfig.json
|       +-- server.json         MCP registry metadata
|       +-- src/
|           +-- index.ts        MCP server entry
|           +-- discovery.ts    discover dockit/sqlkit ports
|           +-- backends.ts     HTTP proxy to each backend
|           +-- tools.ts        unified tool definitions
|
+-- docs/
+-- README.md

Tool naming convention

All tools use a unified data_studio__{db}__{action} format. The database prefix determines which backend handles the call.

data_studio__sql_*       →  sqlkit:9121
data_studio__es_*        →  dockit:9120
data_studio__mongo_*     →  dockit:9120
data_studio__dynamo_*    →  dockit:9120
data_studio__list_*      →  both backends, merge results

The MCP server routes purely by prefix — no need for Agent to know which backend is which.

The MCP server auto-filters tools based on which backends are running:

• Only dockit running → NoSQL tools only (es, mongo, dynamo)
• Only sqlkit running → SQL tools only
• Both running → All tools

Work breakdown

Phase 1: Unified Capability Types (→ #11)

See #11 for details.

Phase 2: HTTP Bridge — Desktop apps as backends (→ #12)

See #12 for details.

Phase 3: MCP Server — TypeScript npm package (→ #13)

See #13 for details.

Phase 4: MCP Permission Model & Management UI (→ #10)

See #10 for full design, research findings, and implementation details.

References

• Model Context Protocol: https://modelcontextprotocol.io/
• @modelcontextprotocol/sdk: https://github.com/modelcontextprotocol/typescript-sdk
• data-studio-agent architecture: ./docs/architecture.md

Dependencies

• dockit and sqlkit must publish a release with MCP bridge before MCP Server can be tested end-to-end

Out of scope

• Replacing the built-in AI assistant with MCP — the in-app agent loop continues unchanged
• JDBC bridge for sqlkit — handled by sqlkit's existing jdbc-bridge/
• Standalone mode (MCP without desktop apps) — can be revisited if there is demand