Skip to main content
Watch: I Built an AI App Without Leaving Databricks

Observability - official Helicone server

Helicone MCP server

Local MCP server that lets AI assistants query Helicone request logs and sessions with filters on model, provider, status, cost, latency and custom properties. It can also send LLM requests through the Helicone AI Gateway. It runs locally from @helicone/mcp (npm) with API key auth.

Last verified 3 tools7 sources

Install

JSON
{
  "mcpServers": {
    "helicone": {
      "command": "npx",
      "args": ["@helicone/mcp@latest"],
      "env": {
        "HELICONE_API_KEY": "sk-helicone-xxxxxxx-xxxxxxx-xxxxxxx-xxxxxxx"
      }
    }
  }
}
TOML
[mcp_servers.helicone]
command = "npx"
args = ["@helicone/mcp@latest"]

[mcp_servers.helicone.env]
HELICONE_API_KEY = "sk-helicone-xxxxxxx-xxxxxxx-xxxxxxx-xxxxxxx"
JSON
{
  "mcpServers": {
    "helicone": {
      "command": "npx",
      "args": ["@helicone/mcp@latest"],
      "env": {
        "HELICONE_API_KEY": "sk-helicone-xxxxxxx-xxxxxxx-xxxxxxx-xxxxxxx"
      }
    }
  }
}

Generic

JSON
{
  "mcpServers": {
    "helicone": {
      "type": "stdio",
      "command": "npx",
      "args": ["@helicone/mcp@latest"],
      "env": {
        "HELICONE_API_KEY": "sk-helicone-xxxxxxx-xxxxxxx-xxxxxxx-xxxxxxx"
      }
    }
  }
}

Tools

query_requests
Query requests with filters, pagination, sorting and optional request/response bodies
query_sessions
Query sessions with search, time range filtering and advanced filters
use_ai_gateway
Make LLM requests through the Helicone AI Gateway with automatic observability

As documented on October 17, 2026. A running server reports its current list through tools/list.

Notes and caveats

  • The Claude Code snippet is the documented .mcp.json / ~/.claude.json config, not a CLI command.
  • API keys come from us.helicone.ai or eu.helicone.ai settings.
  • Latest npm version is 0.1.6 (2025-11-04), with no newer release in almost a year.
  • The docs page lists only query_requests and query_sessions; use_ai_gateway is in the repo README and the Claude Agent SDK guide.
  • The Helicone homepage says Helicone has joined Mintlify.
  • No stage label in the docs.

FAQ

What does the Helicone MCP server do?

Local MCP server that lets AI assistants query Helicone request logs and sessions with filters on model, provider, status, cost, latency and custom properties. It can also send LLM requests through the Helicone AI Gateway. The vendor documents 3 tools, including query_requests, query_sessions and use_ai_gateway.

How do I add the Helicone MCP server to Claude Code?

Helicone documents a Claude Code config; copy it from the Claude Code section of this page. Snippets for Codex, Cursor and Generic are there too.

Does the Helicone MCP server need authentication?

Yes, API key. Transport: stdio. Status: Unlabelled, last checked October 17, 2026.

Sources

Last verified against 7 Helicone pages, then re-checked independently.

Newsletter

New MCP servers, the week they ship

One email per week. New official MCP servers, the coding tools that support them and how developers wire them up.