From a failed request to a fix. In one paste.
ApiTraffic already has the evidence your AI coding agent needs: the failing request, the queries behind it and the deploy that was running. Fix with AI hands it over as a ready-made prompt - and the MCP server lets your assistant go and find it on its own.
A prompt for every failure, slow request and API change
Open any request that returned a 4xx or 5xx and select Fix with AI. ApiTraffic correlates it with your recent deploys and assembles a grounded prompt - the endpoint, the error signature, redacted example requests and the commit most likely to blame - ready for Cursor, Claude Code, Copilot or ChatGPT.
- Ask AI on any waterfall, with every span, its timing and its code location
- Diagnose with AI on any drift finding, with the diff and its baseline
- Commit-aware when you send deploy markers or link a Git repository
Your assistant can read the traffic. Let it do the first pass.
ApiTraffic runs a Model Context Protocol server with 53 tools. Connect Claude, Claude Code, Cursor, VS Code or Windsurf with one URL, approve it in your browser, and ask what broke - your assistant searches requests, diagnoses endpoints, checks a release for regressions and reads the code that was deployed.
- OAuth sign-in - no keys to paste - or a token for CI and automation
- One environment, the buckets you choose, revocable any time
53 tools. No arbitrary access.
A documented, fixed set of tools - this is everything an assistant can reach. A sample from each group:
Diagnosis
diagnose_endpoint, compare_windows, get_error_clusters, get_fix_prompt
Requests
search_requests, get_errors, get_request_detail, replay_request
Drift
list_drift_findings, get_drift_finding, get_drift_prompt, get_drift_health
Performance
slow_operations, operation_trend, regressions_since_deploy, get_trace
Deploys
list_events, create_event, get_deployed_code
Firewall
get_waf_overview, list_waf_matches, get_waf_policy, get_waf_rule
Metrics & reports
get_metrics, get_response_times, run_report, run_saved_report
Workspace & docs
get_workspace_overview, list_endpoints, list_attention_signals, search_docs
One URL. Two minutes.
Add https://api.apitraffic.io/mcp to your client, sign in to ApiTraffic when the browser opens, and pick the environment and buckets it may read.
For Windsurf, Claude Desktop’s config file, scripts and CI, create an MCP token and send it as a bearer header instead.
Settings → Connectors → Add custom connector
Name: ApiTraffic
URL: https://api.apitraffic.io/mcp
Select Connect, then sign in to ApiTraffic.AI access you can hand out without losing sleep
Scoped to what you choose
Every connection is bound to one environment and the buckets you pick. An assistant helping with payments cannot browse your admin API.
Redaction still applies
Fields your team masks are masked for assistants too. The model never sees a card number you already decided not to keep.
Read-mostly by design
No arbitrary queries. The only writes are recording an event marker and preparing a request replay for you to run.
Sign in with OAuth
Paste one URL into your client and approve it in the browser, MFA included. Or use a token for CI and clients that need a static header.
Grounded in production
Prompts carry the real failing requests, the spans and the deploy that was running - so the first suggestion is usually the right one.
Revoke in one click
Connected apps and tokens are listed with when they were last used, and can be disconnected any time.
Let your AI agent read the traffic before you do.
Install the SDK, connect your assistant with one URL, and ask it what broke. Free trial, no credit card.
Free trial · No credit card · Set up in minutes