Level MCP 伺服器的三次呼叫如何查詢柏林招聘 ML 工程師的公司

Three MCP calls that answer "who is hiring ML engineers in Berlin right now?"

Job boards give you a search box. An MCP server gives your assistant the same search box, callable from a conversation. If you run agents locally — Claude Code, Cursor, or anything that speaks MCP — you can stop copy-pasting job URLs into chat and let the agent query the board directly.

Level exposes its AI-rated job catalog through a free, keyless MCP server at https://jobsbylevel.com/mcp (Streamable HTTP, stateless), backed by the same data as the REST API. As of 22 September 2026 the catalog held 47,735 live listings from 481 companies, each rated Level 1 to 4 depending on how central AI is to the actual work.

Wire it up once

claude mcp add --transport http level-jobs https://jobsbylevel.com/mcp

or in any MCP client config:

{ "mcpServers": { "level-jobs": { "url": "https://jobsbylevel.com/mcp" } } }

Call 1 — search with filters

search_jobs(query="machine learning", city="Berlin", ai_level_min=3, remote=1)

search_jobs takes free text plus filters — AI level range, remote flag, city, company — and returns up to 20 results per call. Every result carries a canonical jobsbylevel.com/jobs/<slug> URL; keep that link when the agent shows you results.

Call 2 — pull one full listing

get_job(slug="...")

The listing detail is where the AI rating becomes readable: the 0–100 score, the mapped level, and the posting text the score was computed from. When an agent recommends a role, this is the call that lets you check the recommendation against the source.

Call 3 — ask about the market, not just the listings

market_stats()

Live distribution across the four levels, remote share, and the tools most named in postings. On 22 September 2026 the split was: Level 4 (score 80–100, building AI systems) 3,539 jobs, Level 3 (60–79) 6,574, Level 2 (40–59, AI as a tool) 3,275, Level 1 (0–39) 34,347.

The two remaining tools, list_ai_native_companies and get_level_definitions, cover "which companies actually hire for AI work" and the exact meaning of each level.

Why an agent wants this

The REST equivalent is plain JSON with open CORS — GET /api/v1/jobs?query=machine%20learning&ai_level_min=3&remote=1 — capped at 20 results per page, about 60 requests per minute per IP of fair use. The MCP server wraps exactly the same data, so an agent can loop, filter, and summarize without you hand-writing queries.

Full reference: jobsbylevel.com/developers — free REST API and MCP server, no key, no auth.