AgentOps MCP Server – AI Trace Observability
AgentOps provides an official MCP server that lets AI assistants inspect trace data from an AgentOps project. The server exposes tools for listing recent traces and retrieving trace details so developers can debug agent behavior with live observability context.
Overview
AgentOps is an observability and debugging platform for AI agents and LLM applications. Its official MCP server connects MCP-compatible assistants to the AgentOps API so they can inspect recent trace activity from an AgentOps project while helping a developer troubleshoot an agent run.
The server is intentionally focused on trace access. It is useful when an assistant needs live execution context from AgentOps rather than relying on screenshots, copied logs, or static knowledge about how an agent should behave.
What the MCP server enables
The AgentOps MCP server exposes two documented tools. list_traces lists the most recent traces from an AgentOps project, with optional limiting and support for a custom AgentOps API URL. trace_detail retrieves the details for a specific trace ID returned by list_traces, including associated spans, metadata, timing information, and other raw API response data.
The server handles communication with the AgentOps API and authenticates requests by exchanging an AgentOps API key for a JWT token. It returns raw API responses back to the assistant and does not document write, delete, deployment, metric-management, or project-administration operations.
When to use it
Use this MCP server when an AI assistant is helping investigate an agent execution captured in AgentOps. Practical workflows include asking an assistant to list recent traces, inspect a failing run by trace ID, compare spans inside a trace, or summarize timing and metadata that may explain unexpected agent behavior.
It is especially relevant for teams already instrumenting applications with the AgentOps SDK and using the AgentOps dashboard for observability.
Connection and authentication
The documented local server runs over stdio. AgentOps documents running it through the repository convenience script ./bin/run-server or directly with Python tooling as uv run -m mcp_server_agentops_api. Claude Coder configuration examples point to the local bin/run-server path.
Authentication is supplied to the tools as AGENTOPS_API_KEY on each request. The server can also accept AGENTOPS_API_URL per request for non-production or custom API environments. The production AgentOps API defaults to https://api.agentops.ai.
Key considerations
Treat the AgentOps API key as sensitive project access. The official README states that API keys are not stored between requests and that each request is authenticated by exchanging the API key for a JWT token, but users should still avoid placing secrets in shared prompts, committed configuration files, or logs.
The current documented capability set is narrow: recent trace listing and trace detail retrieval. The README lists additional tools such as metrics, spans, filtering, batch operations, and short-term JWT caching as future enhancements rather than current capabilities.
Supported Transports
stdio
Command: uv
Args:
run-mmcp_server_agentops_api
Frequently Asked Questions
- When should an AI agent use the AgentOps MCP server?
- Use it when the assistant needs to inspect live AgentOps trace data while debugging an AI agent or LLM application that has already been instrumented with AgentOps.
- What does the AgentOps MCP server add to an AI agent's capabilities?
- It adds access to AgentOps trace context through documented tools for listing recent traces and retrieving the detailed spans, metadata, timing information, and raw response data for a selected trace.
- What can the AI agent access or manage through this server?
- The documented server can read recent traces and trace details from an AgentOps project. It does not document write operations, dashboard administration, deployment management, or project configuration changes.
- How is authentication configured for this MCP server?
- The documented tools require an `AGENTOPS_API_KEY` parameter with each request. The server exchanges that API key for a JWT token when communicating with the AgentOps API and does not store the API key between requests.
- Which transport should be used for this MCP server?
- Use the local stdio transport. AgentOps documents running the server with the repository's `./bin/run-server` script or directly with `uv run -m mcp_server_agentops_api`; no hosted Streamable HTTP or SSE endpoint is documented for this server.