Skip to main content

​
Overview

Arize AX offers 2 Model Context Protocol (MCP) servers that enhance your development workflow by integrating Arize support, documentation, and instrumentation guidance directly into your IDE or LLM.

✨Start Here

​
Arize AX Tracing Assistant

Installation →Best for: Instrumentation guidance and implementation supportThe Arize Tracing Assistant provides hands-on help with adding tracing to your applications:
  • Instrumentation Guides: Step-by-step guides for adding Arize AX tracing to your codebase
  • Curated Examples: Framework-specific tracing examples and best practices
  • Direct Support Access: Natural language queries to Arize support directly from your development environment
  • Code Analysis: Analyzes your codebase to identify instrumentation needs and propose implementation plans

​
Arize AX Docs MCP Server

Installation →Best for: Comprehensive documentation access and referenceThe AX Docs MCP Server provides searchable access to the entire Arize AX documentation knowledge base:
  • Full Documentation Access: Search across all Arize AX documentation, guides, and references
  • Code Examples: Find relevant code examples and implementation patterns
  • API References: Access API documentation and endpoint details
  • Feature Guides: Understand how features work and locate implementation details
Best Experience: For the most comprehensive development workflow, use both MCP servers together.

​
Usage

​
Using the AX Tracing Assistant

Once the Tracing Assistant MCP server is running, you can ask your IDE or LLM natural-language questions like:
  • “Instrument this app using Arize.”
  • “Can you use manual instrumentation so that I have more control over my traces?”
  • “How can I redact sensitive information from my spans?”
  • “Can you make sure the context of this trace is propagated across these tool calls?”
For comprehensive instrumentation assistance, use this prompt in your editor’s agent mode with the MCP server enabled: 
Goal: Add Arize AX tracing to my application using the following workflow:

PHASE 1 — ANALYSIS ONLY (READ-ONLY)
- Scan the codebase to identify:
- The primary application language (Python, Java, TypeScript, or JavaScript)
- LLM calls, agent/orchestration frameworks, retrieval/embedding logic
- Existing tracing, OpenTelemetry, or logging instrumentation
- Infer the stack (LLM provider, framework, runtime)
- Identify applicable OpenInference packages based on framework, platform, and/or LLM provider
- When a framework is used, you should also include the OpenInference package for the LLM provider. The framework's OpenInference package does not include tracing for the LLM.
- Identify where manual instrumentation is needed (if any)
- Flag conflicts or overlapping instrumentation
- Propose a minimal, safe instrumentation plan

Rules for Phase 1:
- Do NOT modify code
- Do NOT create or write any files
- Do NOT generate markdown summary files
- Return a single concise summary + implementation plan

PHASE 2 — IMPLEMENTATION (ONLY AFTER PHASE 1)
- Use auto-instrumentation first; manual spans only if required
- Do NOT change business logic or behavior
- Instrumentation must be optional and fail gracefully if env vars are missing
- Create a single centralized instrumentation module (language-appropriate)
- Avoid creating unnecessary files or markdown docs

Deliverables (Phase 2, single response):
- Dependency install commands (pip / npm / pnpm / maven / gradle as applicable)
- Full instrumentation setup code (copy-paste ready)
- Where/how to initialize at application startup
- Required environment variables + .env example
- Minimal before/after code examples
- How to verify traces in Arize AX
- Troubleshooting tips

Behavior:
- Prefer inspection over mutation
- Keep output concise and production-focused
- Do not generate excessive documentation or summary files

​
Using the AX Docs MCP Server

Once the AX Docs MCP server is running, you can ask your IDE or LLM to search the documentation:
  • “How do I set up evaluation for my RAG application?”
  • “Show me examples of tracing with LangChain”
  • “What are the available evaluators in Arize AX?”
  • “How do I configure offline evaluations?”
The server will search across all Arize AX documentation and return relevant content with direct links to documentation pages.

​
Installation

The Arize AX Tracing Assistant is distributed via uv, a fast Python package manager. The AX Docs MCP Server provides HTTP-based access to the Arize AX documentation knowledge base.

​
Install uv (Required for Tracing Assistant only)

📦 View Tracing Assistant on PyPI The Tracing Assistant requires uv. The Docs MCP Server does not require any installation. 
pip install uv
# or
brew install uv

​
Claude Code Integration

You can add the Tracing Assistant directly from the command line:
claude mcp add arize-tracing-assistant uvx arize-tracing-assistant@latest
Alternatively, to add it via JSON:
claude mcp add-json arize-tracing-assistant '{"command": "uvx", "args": ["arize-tracing-assistant@latest"]}'
To verify the server was added correctly:
claude mcp list

​
Gemini CLI Integration

Install using the Gemini CLI extension. (Gemini CLI homepage)
gemini extensions install https://github.com/Arize-ai/arize-tracing-assistant

​
IDE Integration

​
Cursor IDE

  1. Navigate to: Settings → MCP
  2. Click Add new global MCP server
  3. Insert the following into your config JSON: 
    "arize-tracing-assistant": {
  "command": "uvx",
  "args": ["arize-tracing-assistant@latest"]
}

​
Claude Desktop

  1. Open: Settings → Developer → Edit Config
  2. Add the following config: 
    "mcpServers": {
  "arize-tracing-assistant": {
    "command": "uvx",
    "args": ["arize-tracing-assistant@latest"]
  }
}

​
Manual MCP Config

Add this snippet to your mcpServers config section:
"mcpServers": {
  "arize-tracing-assistant": {
    "command": "uvx",
    "args": ["arize-tracing-assistant@latest"]
  }
}

​
Troubleshooting

​
AX Tracing Assistant Troubleshooting

  • Run uv cache clean to reset any cached versions
  • Confirm the uv path using: 
    which uv
  • Start the server manually by running: 
    uvx arize-tracing-assistant@latest
  • Use Anthropic’s MCP inspector: 
    npx @modelcontextprotocol/inspector uvx arize-tracing-assistant@latest

​
AX Docs MCP Server Troubleshooting

  • Verify the server URL is accessible: https://arize.com/docs/mcp
  • Check your network connection if searches are not returning results
  • Ensure your IDE or LLM supports HTTP-based MCP servers