BenchLeader MCP server

Endpoint: https://www.benchleader.com/mcp
Model Context Protocol over Streamable HTTP, stateless JSON responses. Data as of 31 Oct 2026, Index v1.4, refreshed with every daily update.

Connect a client

{
  "mcpServers": {
    "benchleader": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://www.benchleader.com/mcp"
      ]
    }
  }
}

claude mcp add --transport http benchleader https://www.benchleader.com/mcp

{
  "mcpServers": {
    "benchleader": {
      "url": "https://www.benchleader.com/mcp"
    }
  }
}

{
  "servers": {
    "benchleader": {
      "type": "http",
      "url": "https://www.benchleader.com/mcp"
    }
  }
}

{
  "mcpServers": {
    "benchleader": {
      "serverUrl": "https://www.benchleader.com/mcp"
    }
  }
}

https://www.benchleader.com/mcp

Tools

• get_leaderboard: The BenchLeader Index leaderboard of large language models: a daily composite of independent public benchmarks, one row per model (its best reasoning-effort configuration) as on the homepage. Optionally rank by one category score (coding, agentic, reasoning, math, knowledge…), list every model × effort configuration, or filter to open-weights models or one developer. Paged.
• search_models: Find models by name, alias, id or developer ("opus", "gemini 4", "deepseek"). Returns each match's id, rank, Index and reasoning-effort configurations, for use with get_model and compare_models.
• get_model: One model's BenchLeader profile: Index and rank, category scores, list price, speed, latency, context window, its other reasoning-effort configurations, and its strongest independent benchmark results (each with its board rank, publisher and source link).
• compare_models: Head-to-head comparison of two models with BenchLeader's verdict: who leads on quality (a lead is claimed only outside the combined margin of error), price and speed ratios, category-by-category wins, when to choose each, and shared benchmark results. Links the comparison page.
• list_benchmarks: The benchmarks BenchLeader reads: what each measures, publisher, category, whether it feeds the Index (and its weight), how many models it covers, status and caveats. Filter by category or a search query. Paged.
• get_benchmark: One benchmark: what it measures, unit, publisher and source, whether and how much it counts in the BenchLeader Index, status and caveats, and its top models (one row per model, best result first).
• about_benchleader: What BenchLeader is, how the BenchLeader Index is computed, the current method version, data date and update cadence, coverage counts, licence and how to cite it.

Every result carries the canonical BenchLeader URL of the model, comparison or benchmark page it summarises, plus the publisher’s source link for each benchmark result. Lists are paged: the leaderboard returns at most 100 rows per call, benchmark lists 100, a model’s results 50 and a benchmark’s top models 25.

Try it from a terminal

curl -s https://www.benchleader.com/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"get_leaderboard","arguments":{"limit":5}}}'

Send initialize and tools/list the same way. A GET on the endpoint with Accept: application/json describes the server.

Terms

The answers are the same figures the site shows, computed by the method on the methodology page. BenchLeader's own computed figures (the BenchLeader Index, its interval, the category scores and the rank) are licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/): commercial and non-commercial use alike, with attribution — credit "BenchLeader" with a link to https://www.benchleader.com (or to the row's url). That licence does not cover anyone else's figures and BenchLeader does not relicense them: list prices are the vendors' own published facts, and speed, latency and the benchmark results behind the scores remain under their publishers' own terms (credited on each model page and listed under "sources"; Epoch AI's Benchmarking Hub and DuelLab GameBench 2, for example, are CC BY 4.0).

The server answers questions; it is not a bulk download. For the published dataset see open data.