TracePulse

Get recent error and warning events sorted by signal_score descending.

Get recent log events at any severity level sorted by timestamp descending.

Quick health check - connection state, error count, last error time.

Clear events from the buffer. Pass fingerprint to clear a specific error, or omit to clear all.

Mark an error as investigated. Acknowledged errors are excluded from get_errors results. Saves tokens by preventing re-investigation.

Watch N seconds for new errors after a code change. Returns events + hot_reload_detected.

Get current build/compilation errors (TypeScript, ESLint, Vite/webpack). Returns only errors with source 'build-error'.

Group errors by type + module path. See patterns across the codebase.

Check or apply database migrations. Auto-detects alembic/prisma/django/knex. Pass apply=true to run pending migrations.

Review your own MCP tool usage this session. Shows which tools were called, when, with what params, and response sizes. Use to optimize your workflow.

Analyze agent effectiveness: uninvestigated errors, verification gaps, tool usage patterns, parser stats. Use at end of session.

Check for env, dependency, and migration drift in one call. Detects missing .env vars, lock files, and migration frameworks.

Environmental report: tokens saved, energy saved (Wh), CO2 saved (g). Shows the impact of using TracePulse this session.

Compact session manifest: errors seen/acknowledged/pending, build status, tool count. ~200 tokens. Use after context compaction.

Per-endpoint response time percentiles (P50, P95, max) from HTTP access logs. Use to detect performance regressions after code changes.

Deep-dive into a specific error by fingerprint. Returns the full error, surrounding logs (±5s), and occurrence count.

Get a unified chronological stream of ALL events in a time window. Use for full situational awareness.

List all monitored services with status, error count, and last activity.

Match browser HTTP failures with backend stack traces. Returns paired errors with confidence scores.

Get only errors with fingerprints not seen in previous sessions. Use `since` (Unix ms) to scope to a specific time window — e.g., pass the timestamp from when a smoke test started to see only errors that appeared during that test run.

Cross-session frequency and history for a specific error fingerprint.

Link recent errors to uncommitted git changes. Shows which errors may be caused by your recent edits.

Cross-layer failure diagnosis. Correlates backend logs, frontend errors, git state, and process state to identify root causes that span multiple layers.

One-line health check: error count, warning count, total events, uptime. Use instead of calling get_runtime_status + get_errors + get_server_logs separately.

Post-fix check: watch + build + errors in one call. Returns pass/fail verdict.

Composite fix verification: checks new errors, pinned fingerprint resolved, build clean, HMR detected. Returns confidence-scored verdict (high/medium/low). Collapses 5-7 calls into 1.

Pre-assembled reasoning packet for an error: message + stack + surrounding logs + file context. Token-budgeted. One call instead of 4-5.

Type-check + build + runtime error check in one call. Replaces 3 separate tool calls.

Block until next build/hot-reload completes. Event-driven, no polling.

Block until the next event of a specific type arrives. Types: error, warning, build, crash, any.

Run a command, parse output through 26 parsers, return structured pass/fail results. Use INSTEAD OF shell for tests, builds, and linters. If timeout occurs, increase timeout_seconds (up to 120). Never fall back to shell for commands that produce pass/fail output.

Get recent HTTP requests from access logs. Filter by URL path and status code. Shows method, path, status, and timing.

Kill and respawn the dev server process. Only works in start mode. Use after installing dependencies, changing config, or when the server is stuck.

Summary of all discovered backend services (databases, Redis, queues) with connectivity status. Reads from .env files, probes every 60s.

Detailed status for a specific infrastructure service including connection history. Use after get_infra_status to investigate an unreachable service.

Check if TCP port(s) are available or in use on localhost. Accepts single port or array.

Kill the process occupying a port. Use when start_server fails because a port is in use from a crashed previous session. Do not use shell with lsof/kill — this tool does it safely.

Composite health check: server status + infrastructure connectivity + error count + build status in one call. Use as the first call in any debugging session.

Test that an MCP server starts and responds to the initialize handshake. Sends the same JSON-RPC message that Kiro/Claude/Cursor sends on connect. Use after changing MCP server code, imports, or dependencies.

Register a health probe for a critical endpoint. TP will check it periodically and alert on failure. Use after building a new API route.

List all registered health probes with their latest results (pass/fail/error).

Analyze cross-session error patterns: recurring bugs, velocity changes, error chains, flaky errors, regressions. Requires persistence (default).

Start a dev server for live error monitoring. Pre-validates the command and returns diagnostics if invalid. Use when TracePulse started without a server command. On failure, use run_and_watch with the same command to see full error output — do not fall back to shell.

Stop a running dev server. Sends SIGTERM, waits, then SIGKILL if needed.