Inspect recorded agent runs, span trees, slow tools and approval logs with abhishekash’s read-only MCP Trace server.
MCP Trace lets an assistant inspect existing OpenTelemetry-style JSONL traces. It exposes run summaries, nested span trees, slow spans and approval records for investigating what happened during an agent run.
Start with a run identifier, inspect the span tree and compare the longest spans with the overall duration. An approval log can show where human decisions occurred. Keep the trace identifier and relevant spans with the explanation so a developer can inspect the same evidence.
The package name is abhishekash-mcp-trace, although the project is called mcp-trace. Launch it with uvx abhishekash-mcp-trace --trace-dir /path/to/traces. The server reads trace files; it does not automatically instrument an application or create traces that were never recorded. The documented initial tool surface is read-only.
Check the linked publisher documentation and your connected client's tool list. Names and availability can vary with the server version and account permissions.
| Tool | What it does |
|---|---|
| list_runs | List recorded runs. |
| run_summary | Inspect one run. |
| span_tree | View the nested spans in a run. |
| slowest_spans | Find spans with the longest durations. |
| approval_log | Inspect recorded approval decisions. |
{
"mcpServers": {
"agent-traces": {
"command": "uvx",
"args": [
"abhishekash-mcp-trace",
"--trace-dir",
"/path/to/traces"
]
}
}
}Replace the directory with existing compatible trace files.
Python and uv for uvx, plus a directory containing compatible recorded trace files.
Give your coding agent the full DevTools toolbox: traces, network, console, heap snapshots and Lighthouse.
Workers, KV, R2 and D1 by conversation — Cloudflare's official remote servers for building and observability.
A powerful MCP (Model Context Protocol) server that gives AI agents web superpowers. Extract content, search the web, monitor sites, and convert HTML to markdown — all through a clean tool interface.
This project uses [semver](https://semver.org/) for release versions. We have not graduated to 1.0, so breaking changes may occur for minor version
MCP server for monitoring remote file sources, detecting new files, and extracting content
Independent, evidence-backed reviews of AI agents. C2PA-signed proof, machine-readable scorecards, and a Laddoo Score that's consistent across every reviewed agent — so buyers, teams, and other…