MCP 호환 에이전트 및 툴링에 Databricks REST 기능을 노출하는 Production-ready ** Model Context Protocol (MCP)** 서버. 버전 0.4.4는 구조화된 응답, 리소스 캐싱, 리트리웨어 네트워킹 및 엔드 투 엔드 탄력 개선을 소개합니다.

---

내용 표

1. 핵심 기능
2. 건축 특징
3. 임명
4. 윤곽
5. 서버 실행
6. MCP 클라이언트와 통합
7. 공구 응답으로 일
8. 유효한 공구
9. 발달 Workflow
10. 테스트
11. 출판물
12. 지원 & 접촉
13. 라이센스

---

키 기능

• Structured MCP Responses - 각 도구는 CallToolResult] content]] 및 structuredContent]]에서 기계 읽기 쉬운 페이로드를 사용하여 도구의 outputSchema를 준수합니다.
• **출처 캐싱 ** - 대형 노트북 / 워크스페이스 수출은 한 번 캐쉬하고 resource_link]] resource://databricks/exports/{id}]과 같은 URI와 콘텐츠 블록 (편의 메타 데이터에 반영). -Progress & Metrics - 긴 실행 작업 흐름 MCP 진행 알림 및 트랙 per-tool 성공/error/timeout/cancel 메트릭.

Databricks MCP Server

A production-ready Model Context Protocol (MCP) server that exposes Databricks REST capabilities to MCP-compatible agents and tooling. Version 0.4.4 introduces structured responses, resource caching, retry-aware networking, and end-to-end resilience improvements.

---

Table of Contents

1. Key Capabilities
2. Architecture Highlights
3. Installation
4. Configuration
5. Running the Server
6. Integrating with MCP Clients
7. Working with Tool Responses
8. Available Tools
9. Development Workflow
10. Testing
11. Publishing Builds
12. Support & Contact
13. License

---

Key Capabilities

• Structured MCP Responses - Each tool returns a CallToolResult with a human-readable summary in content and machine-readable payloads in structuredContent that conform to the tool’s outputSchema.
• Resource Caching - Large notebook/workspace exports are cached once and returned as resource_link content blocks with URIs such as resource://databricks/exports/{id} (also reflected in metadata for convenience).
• Progress & Metrics - Long-running actions stream MCP progress notifications and track per-tool success/error/timeout/cancel metrics.
• Resilient Networking - Shared HTTP client injects request IDs, enforces timeouts, and retries retryable Databricks responses (408/429/5xx) with exponential backoff.
• Async Runtime - Built on mcp.server.FastMCP with centralized JSON logging and concurrency guards for predictable stdio behaviour.

Architecture Highlights

• databricks_mcp/server/databricks_mcp_server.py - FastMCP server with tool registration, progress handling, metrics, and resource caching.
• databricks_mcp/core/utils.py - HTTP utilities with correlation IDs, retries, and error mapping to DatabricksAPIError.
• databricks_mcp/core/logging_utils.py - JSON logging configuration for stderr/file outputs.
• databricks_mcp/core/models.py - Pydantic models (e.g., ClusterConfig) used by tool schemas.
• Tests under tests/ mock Databricks APIs to validate orchestration, structured responses, and schema metadata without shell scripts.

For an in-depth tour of data flow and design decisions, see ARCHITECTURE.md.

Installation

Prerequisites

• Python 3.10+
• uv for dependency management and publishing

Quick Install (recommended)

Register the server with Cursor using the deeplink below - it resolves to uvx databricks-mcp-server@latest and picks up future updates automatically.