agents-libagents-lib
AgentsMCP ServersWorkflowsTutorialsComparisonsTemplatesNewsAbout
🔌

Helicone MCP

Observability•⭐ 4,200

Open-source LLM gateway for observability, caching, and rate limiting across all providers.

GitHubWebsite →
Claude DesktopCursorWindsurf

Overview

Open-source LLM gateway for observability, caching, and rate limiting across all providers.

Setup

Run with npx:

npx -y @modelcontextprotocol/server-helicone

Configuration

HELICONE_API_KEY environment variable

Documentation

Helicone MCP

Overview

Helicone MCP is a Model Context Protocol server that provides integration with Helicone, an open-source LLM gateway for observability, caching, and rate limiting across all LLM providers. It enables AI agents to monitor and manage their LLM usage through a unified interface.

Helicone acts as a middleware between your application and LLM providers, providing comprehensive analytics, cost tracking, and performance optimization without requiring changes to your code.

Features

  • LLM Observability — Track all LLM requests and responses
  • Cost Tracking — Monitor spending across all LLM providers
  • Request Caching — Cache responses to reduce costs and latency
  • Rate Limiting — Control request rates per provider
  • Provider Routing — Route requests to different providers
  • Anonymization — Anonymize sensitive data in logs
  • Self-Hosted Option — Deploy your own Helicone instance

Installation

npx -y @modelcontextprotocol/server-helicone

Configuration

{
  "mcpServers": {
    "helicone": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-helicone"],
      "env": {
        "HELICONE_API_KEY": "your-api-key"
      }
    }
  }
}

Available Tools

ToolDescription
get_requestsList recent LLM requests
get_request_detailsGet details of a specific request
get_usage_statsGet usage statistics
get_cost_breakdownGet cost breakdown by provider
get_cache_statsGet cache hit/miss statistics
get_propertiesGet custom properties set on requests

Usage Examples

Get Recent Requests

requests = helicone.get_requests(
    limit=50,
    offset=0
)
for req in requests:
    print(f"{req.model}: {req.total_tokens} tokens")

Get Cost Breakdown

costs = helicone.get_cost_breakdown(
    start_time="2026-05-01T00:00:00Z",
    end_time="2026-05-31T23:59:59Z"
)
for provider, cost in costs.items():
    print(f"{provider}: ${cost}")

Cache Statistics

cache = helicone.get_cache_stats()
print(f"Cache hit rate: {cache.hit_rate:.2%}")
print(f"Cost saved: ${cache.cost_saved}")

Claude Desktop Setup

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "helicone": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-helicone"],
      "env": {
        "HELICONE_API_KEY": "your-api-key"
      }
    }
  }
}

Self-Hosted Option

Helicone can be self-hosted for full data control:

docker run -d \
  --name helicone \
  -p 8585:8585 \
  -e HELICONE_API_KEY=your-key \
  helicone/helicone:latest

Pros

  • ✅ Open-source and self-hostable
  • ✅ Framework-agnostic (works with any LLM client)
  • ✅ Comprehensive cost tracking
  • ✅ Built-in caching for cost savings
  • ✅ Provider-agnostic routing
  • ✅ Privacy-focused with anonymization

Cons

  • ❌ Requires Helicone account or self-hosting
  • ❌ Additional hop in request path
  • ❌ Some advanced features require paid tier
  • ❌ Setup complexity for self-hosted option

When to Use

Helicone MCP is ideal for:

  • Multi-provider LLM applications
  • Cost-conscious applications needing detailed tracking
  • Teams wanting self-hosted observability
  • Applications needing request caching
  • Privacy-focused deployments

Consider alternatives when:

  • You need deep LangChain integration (use LangSmith)
  • You want zero-code setup (use AgentOps)
  • You only need simple cost tracking (use provider dashboards)

Resources