MCP spec v2 drops Jul 28 · Is your server ready?

MCP servers
break silently.
Yours won't.

Right now your MCP server is either passing or failing calls. You just don't know which. One command wraps it — reliability score in 30 seconds, dangerous call alerts before your users notice, spec v2 readiness check included. Zero code changes.

$npm install -g @mcppulse/cli

$mcp-pulse -- npx @modelcontextprotocol/server-postgres

mcp-pulse -- wraps your existing server · any stdio MCP server

Prove it's healthy:

MCP Pulse84 / 100 · Stable

industry median: 61

Open dashboard →See how it works ↓

0 ms

latency overhead

3,241+

servers benchmarked

industry median score

dashboard.mcp-pulse.com

live

Last 7 days

Servers

server-postgres

84 / 100

server-github

71 / 100

server-slack

58 / 100

Reliability score

6-factor composite

Success rate

97.3%

243 passed · 7 failed

+0.5% this week

p95 latency

847ms

p50: 142ms · 6× spread

+12% this week

Dangerous calls

LIVE

3this week

1 unresolved

delete_table2m

–2 vs last week

Schema drift2 events

server-slack

Spec v22 / 4 ✓

Jul 28 cutover

Auth postureAPI key

score: 50/100

Week over week

Calls387→462

Errors28→41

p95612ms→847ms

Score81→84

AI root cause

Burst: 47 calls / 90s

claude-3-haiku loop — add rate limit client-side

AI Report + Email AlertsStarter

Active policies

BLOCKdelete_*

APPROVEadmin_*

ALLOWread_*

Audit log

✗ BLOCKEDdelete_table2m

✓ ALLOWEDread_schema4m

Governance + Audit LogSecurity

Team scores

4 servers

Security

Platform

Data

ML Ops

⚠ ML Ops below threshold

Score 61 · target ≥ 70

Team Dashboard + BenchmarksTeam

How it works

Three commands.
Full visibility.

Before

User files a bug report

Hours in raw JSON logs

No baseline to compare

spec v2 blind spot

→

After

Alert before users notice

AI root cause in seconds

Score history + benchmark

v2 compat per session

Install

One npm command. Works with npx servers, Node.js processes, Docker containers, and systemd services.

npm install -g @mcppulse/cli

Wrap

Prefix your existing server command with mcp-pulse --. No config files, no SDK changes.

mcp-pulse -- npx @modelcontextprotocol/server-postgres

See everything

Open the dashboard. Reliability score, latency trends, schema drift, and dangerous call alerts — live on the next call.

open http://localhost:3002

Already using Claude Desktop or Cursor? mcp-pulse init auto-wraps every server in your config file.

Running in Docker? CMD ["mcp-pulse", "--", "node", "server.js"] — one line change in your Dockerfile.

Features

Everything your MCP stack
needs to ship with confidence

From local dev to production governance — one wrapper covers it all.

⚡

Zero code changes

Prefix any MCP command with mcp-pulse --. Works with npx servers, Docker containers, and systemd services. stdout is byte-for-byte identical — verified by CI on every commit.

🎯

SLO for your AI stack

The 0–100 reliability score is the SLO you've always wanted for MCP. Six weighted dimensions: success rate, latency stability, schema consistency, error resilience, auth posture, spec v2.

🔍

Schema drift detection

Baselines every tool's param and response shape. Alerts the moment a server update silently breaks the contract — before your users file a bug.

🚨

Cut MTTR from hours to minutes

AI root cause analysis identifies exactly which method, which caller, and which change caused the failure. Email + Slack on dangerous calls. Post-mortem data ready instantly.

🛡️

Governance layer

Block methods by pattern, require human approval before execution, export audit logs to SIEM. PagerDuty and OpsGenie webhooks for on-call routing. Security tier.

🧪

Spec v2 ready

Detects every July 28 2026 breaking change. Stateless sessions, _meta field, error code migration — all checked per session automatically.

Export everywhere

Your data, your existing stack

✓ Slack✓ PDF report✓ PagerDuty✓ OpsGeniePrometheus · Grafana · Datadog — on roadmap

Unlocked from Starter ($9/mo)

The right chart
for every metric.

Each panel is designed for how that specific data is actually read by humans — not just what fits in a grid.

MCP Pulse — server-postgres

Last 7 dayslive

Reliability score

Stable

6-factor weighted
composite

Score breakdown — sorted worst first

Auth posture

Latency stability

Error resilience

Success rate

Schema consistency

Spec v2 compat

100

Success rate

97.3%

✓ 243 passed✗ 7 failed

7-day trend

Latency

142ms

p50 median

847ms

p95 worst

p95 over 7 days

Schema drift

2 events

query2h ago

new field: cursor

insert_row1d ago

removed: batch_id

Dangerous calls

3 alerts

delete_tabledelete2m ago

execute_scriptexecute14m ago

send_messagesend1h ago

Spec v2 compat

Cutover: July 28 2026

✓Stateless sessions

✗_meta on every request

✓Error code -32602

✗Required headers

Auth posture

✓

No auth

API keycurrent

OAuth 2.1

Upgrade to OAuth 2.1 for +10pts

🔒Community benchmarksTeam tier

🔒Governance policiesSecurity tier

🔒Audit export + RBACEnterprise tier

Live at dashboard.mcp-pulse.com — free tier sees score + summary · Starter unlocks the full dashboard

Data & Privacy

What leaves your machine?

The wrapper sits between your AI agent and your tools. Here is exactly what we see — and what we don't.

🖥️

Stays on your machine

All JSON-RPC frames
Actual param values
Response payloads
Session JSONL files
SQLite database
API keys and credentials

☁️

Synced (paid tiers only)

Anonymous reliability score
Method names (not values)
Param key shapes only
Aggregated latency p95
Pass / fail counts
Never tied to your identity

🚫

Never collected

Actual tool call values
Database query results
File or secret contents
User identity or email
IP address
Any PII

Free tier: 100% local. No outbound requests.

Everything lives in ~/.mcp-pulse/ — SQLite database and JSONL session files. Paid tiers sync only anonymous benchmark scores. Your actual tool call data never leaves your machine.

Read the source →

Pricing

Start free. Scale when it matters.

No credit card. Your data stays local. 30 seconds to your first score.

MonthlyAnnual 2 months free

Free

500 calls / mo

$0forever

Know if your server is healthy before your users do.

Install free

✓500 monitored calls / month
✓7-day data retention
✓Reliability score 0–100
✓Terminal session summary
✓Schema drift detection
✓JSONL fallback durability
✓mcp-pulse init (auto-wrap)

Starter

10k calls / mo

$29/mo

Stop debugging from cold JSON logs. Root cause in seconds, not hours.

Get Starter

✓10,000 monitored calls / month
✓30-day history
✓Live web dashboard
✓PDF + HTML reports
✓AI fix suggestions (Claude Haiku)
✓Email alerts — score drop
✓Email alerts — dangerous calls
✓README badge (badge.mcp-pulse.com)

MCP serversbreak silently.Yours won't.

Three commands.Full visibility.

Install

Wrap

See everything

Everything your MCP stackneeds to ship with confidence

Zero code changes

SLO for your AI stack

Schema drift detection

Cut MTTR from hours to minutes

Governance layer

Spec v2 ready

The right chartfor every metric.

What leaves your machine?

Start free. Scale when it matters.

Questions developersactually ask.

MCP servers
break silently.
Yours won't.

Three commands.
Full visibility.

Everything your MCP stack
needs to ship with confidence

The right chart
for every metric.

Questions developers
actually ask.