observability
MCP-native analytics, not page views
Tool calls, sessions, error rates and latency percentiles broken down by client, client version, protocol version and country, with discovery traffic excluded by default.
Our open source tools are used by developers at top companies
How it runs
Understand how your server is actually used
- 1
Measure
Every call captured at the gateway.
- 2
Segment
Split by client, version and region.
- 3
Diagnose
Trace errors to a single session.
- 4
Alert
Get told before users complain.
Per-tool latency
p50, p95 and p99 for every tool, resource and prompt, so you know which one is slow.
last 24h ↑ 12.4%
By country34 regions
Tracesp50 42ms
- tools/call display_weatherclaude142ms
- resources/listchatgpt18ms
- tools/call search_docscursor842ms
- prompts/listcopilot12ms
- tools/call fetch_urlclaude503
p50
38ms
p95
112ms
errors
0.2%
Session replay
The full timeline of a conversation, including the arguments each tool received.
San Francisco, CA
72°F
Partly cloudy · H 74° L 58°
→ tools/call
{
"name": "display_weather",
"arguments": { "city": "San Francisco" }
}← result20041ms
{ "content": [{ "type": "resource" }] }GPT-5.2
18/18
Claude
18/18
Gemini
17/18
Issue triage
Errors grouped into issues with first-seen, last-seen and affected clients.
- Protocol & discovery
- Tool conformance
- Security & policy
- !Domain / TLS / CSP
Submission pack
In the box
Numbers that mean something for MCP
Tool-call volume
What the models actually reach for.
Client breakdown
Which hosts your users are on.
Error rates
Grouped, deduplicated and triaged.
Internal traffic
Dashboard calls excluded by default.
Stop guessing how your MCP server behaves
Explore the rest of the platform
Everything on mcpfy Cloud shares one pipeline.