gekro
GitHub LinkedIn
AI

MCP / Agent Trace Visualizer

Paste a Claude Code, Cursor, Anthropic, or OpenAI trace - see the agent's tool calls as a collapsible tree with categories, results, and failure flags.

Paste trace

No trace yet.

Trace stats

Format—
Tool calls0
Unique tools0
Failures0
Assistant turns0

Tool frequency

  1. — no calls —

Call tree

Paste a trace on the left, or load a sample.

Supported formats + tool color legend

Anthropic SDK — paste the entire messages array, or a full response object with {"messages":[...]}. Tool calls appear as content blocks with type:"tool_use"; results are type:"tool_result" in subsequent user turns.

OpenAI — paste the chat completion with messages containing tool_calls arrays on assistant turns and matching tool role messages for results.

Claude Code JSONL — one JSON object per line (the format Claude Code writes to ~/.claude/projects/.../*.jsonl). Each line is a turn.

fs Read, Write, Edit, Glob, MultiEdit, ls, cat exec Bash, shell, exec, run, sh search Grep, search, find web WebFetch, WebSearch, fetch, http mcp mcp__* prefixed meta TodoWrite, TaskCreate, internal other anything else

As-is, no warranty. These apps are free under their listed license and run entirely in your browser. Use at your own risk — don't blame me if your PC catches fire, your dog runs away, or the math turns out wrong. Verify anything that actually matters. None of this is professional financial, medical, legal, or engineering advice.

© 2026 Rohit Burani · MIT · Built at gekro.com · View source ↗

Guide

What It Does

Paste a trace. See the agent’s tool calls as a collapsible tree.

  • Format auto-detection — Anthropic SDK messages, OpenAI tool_calls, or Claude Code JSONL
  • Categorized tool pills — fs / exec / search / web / mcp / meta / other (heuristic name match)
  • Pass / fail flags — green ✓ for successful calls, red ✗ for tool_results marked with is_error: true
  • Input + result inspection — click any call to expand JSON; toggle “Show results” to see the matched tool_result inline
  • Stats panel — total calls, unique tools, failures, assistant turns, top-8 most-called tools
  • Three samples — one per supported format, so you can see what it does before pasting your own
  • Export — download the normalized call list (turn index, tool name, category, input, result, is_error) as JSON

When To Use It

  • Reviewing a long Claude Code session before sharing with a teammate
  • Debugging “why did the agent loop on this for 6 calls?” - the tree makes the pattern obvious
  • Comparing an OpenAI Assistants run against a Claude run on the same task
  • Auditing MCP tool usage for security review (which mcp__ tools fired, in what order, against what inputs)
  • Building a screenshot for a postmortem or LinkedIn post about an agent run

What’s NOT Included

Why this and not n8n

We deliberately are NOT a visual workflow designer. n8n, Zapier, and Make own that space and they’re adding MCP support. This is a read-only debugging visualizer for traces that already happened. Different job, no overlap.

For informational purposes only. Not financial, medical, or legal advice. You are solely responsible for how you use these tools.

← All apps