Webhook tracing for AI agents

Trace every event and tool call behind your AI agents.

Capture, replay, and correlate the webhooks, callbacks, and tool calls behind your AI workflows — in real time, so agentic systems never fail silently in production.

Get started freeRead the docs
agents · toro · livestreaming
Events
3,901
p95
2.5s
Spend
$27.62
Tool calls
1,847
Volume / hourlast 24h
  • llm_turn$0.0086—
  • exec_command83ms
  • llm_turn$0.0090—
  • http_fetch2341ms
Approve

Hold every tool call back from production until you say go.

Gate destructive actions behind named approval rules. Agent can browse, draft, and reason freely — but writes, refunds, and deletes pause for a human until someone hits approve.

  • Per-tool and per-argument rules — approve once, deny forever
  • Slack, email, and webhook prompts with full call context
  • Audit trail of every decision, who made it, and why
See approval rules →
approvals · toro1 pending
!Awaiting approval
Rule: write to prod
db.write_orders· called by toro
{
  "table":  "orders",
  "id":     "ord_8H2k…",
  "patch":  { "status": "refunded",
              "amount":  -49900 }
}
expires in 4m 12s
Recent decisions
  • db.write_ordersapprovedleah@2m
  • http.deletedeniedleah@14m
  • fs.readautorule#321m
  • stripe.refundapprovedmo@38m
Spend

Track cost per agent. Catch a runaway before it costs you a weekend.

Live cost meters across models, tools, and tenants. Set hourly and monthly thresholds — when burn rate breaks them, we page you before the bill does.

  • Per-token cost attribution across OpenAI, Anthropic, and self-hosted
  • Hourly, daily, and monthly thresholds with Slack + PagerDuty alerts
  • Stop runaway loops automatically when cost gates trip
See cost controls →
llm spend · torothreshold breached
Spend (24h)
$18.01
Burn / hr
$0.84
Threshold
$2.00 / hr
Cost / hourlast 24h
$2.00 / hr
!
Burn rate exceeded threshold
spent $3.79 in the 12:00–13:00 window · alert sent to #oncall
2m ago
gpt-5.5$0.62
56 calls62%
claude-4.7$0.38
29 calls38%
Install

Two commands. Live telemetry. No SDK rewrite.

ScoutTrace is a local, open-source MCP proxy that byte-tees the JSON-RPC traffic between your host and its servers. Wire it in once — every tool call, latency, and cost flows into your Agents view, with redaction running locally before anything leaves the laptop.

  • One Homebrew tap, one init — Claude Code, Cursor, Desktop, any MCP host
  • Strict / standard / permissive redaction runs on-device before dispatch
  • Apache-2.0 on GitHub — point it at WebhookScout, your own webhook, or NDJSON
ScoutTrace on GitHub →
~/projects/toro · zshscouttrace · v0.5.2
$brew install applexica/tap/scouttrace
==> Pouring scouttrace--0.5.2.bottle.tar.gz
🍺 /opt/homebrew/Cellar/scouttrace/0.5.2
$scouttrace init
✓ stored API key in macOS keychain
✓ default destination · webhookscout
$scouttrace claude-hook install --scope local
✓ patched .mcp.json · backup saved
✓ ready — open Claude Code to start capturing
live capture
↑llm_turngpt-5.5113.3k in · $0.0086
↑exec_commandshellcwd /repo · 83ms
↑http_fetchGETstripe · 429 · retry
$
Capture

Inspect every webhook your providers send.

See full headers, payloads, and signature checks at the moment they arrive — no more digging through provider dashboards.

  • Full request + response capture
  • Signature validation for Stripe, GitHub, Slack, and more
  • Forward to your local dev box with one command
Explore the inspector →
webhookscout · live
  • 200POST/webhooks/stripe142ms
  • 200POST/webhooks/github88ms
  • 500POST/agents/run2341ms
  • 200POST/webhooks/slack61ms
  • 200POST/webhooks/shopify116ms
Replay

Replay any captured event, before you ship.

Test your handler against the exact bytes your provider sends. Diff responses across versions to catch regressions.

  • One-click replay
  • Bulk replay suites
  • Diff against previous runs
See replay in action →
webhookscout · live
  • 200POST/webhooks/stripe142ms
  • 200POST/webhooks/github88ms
  • 500POST/agents/run2341ms
  • 200POST/webhooks/slack61ms
  • 200POST/webhooks/shopify116ms
Correlate

See the full chain across services.

When a webhook kicks off three downstream jobs, you see all of them in one timeline — across queues, agents, and tool calls.

  • Cross-service trace IDs
  • Agent + tool-call linking
  • Latency and error attribution
How correlations work →
webhookscout · live
  • 200POST/webhooks/stripe142ms
  • 200POST/webhooks/github88ms
  • 500POST/agents/run2341ms
  • 200POST/webhooks/slack61ms
  • 200POST/webhooks/shopify116ms
Everything you need

A complete webhook surface, in one tool.

Inspector

Live capture for every endpoint with full payload search.

Replay

Re-fire any captured request to your handler.

Schemas

Validate provider payloads against built-in or custom contracts.

Forwarding

Pipe captured webhooks to your local dev box.

Alerts

Slack, Discord, email when forwarding fails or quotas hit.

Cron

Watch scheduled jobs and miss-detect silent failures.

Included with the platform.

Free tier

Start tracing webhooks at no cost.

Real-time

Sub-100ms capture and inspector updates.

Validation

Schema checks for major providers built in.

Integrations

Stripe, GitHub, Slack, Shopify and more.

Get going in 30 seconds

One command. Live endpoint. Real traffic.

Spin up a tracked endpoint and start sending events. Replace the URL anywhere your provider posts.

bash
# Create a tracked endpoint
curl -sX POST https://api.webhookscout.com/v1/endpoints \
  -H "Authorization: Bearer $WHS_KEY" \
  -d '{"name": "stripe"}'
# → https://webhookscout.com/u/abc123

Trace your first webhook in 30 seconds.

Free to start. No credit card. Replay your first capture before you finish your coffee.

Get started freeView pricing