# AgentPulse > Real-time ground-truth hub and breaking-change oracle for autonomous AI agents, coding assistants (Claude Code, Cursor, Cline), and CI bots. Verifies API availability, active 5xx incidents, and SDK schema deprecations in < 150 tokens with sub-50ms latency before bots burn tokens retrying broken endpoints. ## Core API & Discovery - [OpenAPI 3.1 Specification](http://agentpulse-43di.onrender.com/openapi.json): Strict machine-readable API definitions for autonomous tool calling. - [Full Agent Context Documentation](http://agentpulse-43di.onrender.com/llms-full.txt): Complete operational guide, payload schemas, and client retry policies. - [Self-Serve Bot Onboarding](http://agentpulse-43di.onrender.com/v1/agents/register): Zero-CAPTCHA programmatic API key issuance (100 requests/day free tier). ## Essential Preflight & Status Tools - [Batch Preflight Check](http://agentpulse-43di.onrender.com/v1/preflight): POST dependency list (e.g. ["stripe", "supabase", "resend"]) to obtain immediate pass/degraded/block verdict. - [Service Status Query](http://agentpulse-43di.onrender.com/v1/status/stripe): GET compact health status, active incidents, affected scope, and prescriptive agent_hint (< 150 tokens). - [Delta Feed of Breaking Changes](http://agentpulse-43di.onrender.com/v1/changes): GET recent SDK deprecations and schema mutations since timestamp checkpoint. Add `?scope=mine` (auth required) to restrict the feed to your manifest services. Every change carries a `playbook_url`. - [Migration Playbook](http://agentpulse-43di.onrender.com/v1/changes/{id}/playbook): GET a structured migration playbook for one breaking change: `{ change_id, service, kind, summary, steps: [{ action, detail, code_example? }], effort_estimate, risk_notes }`, generated from the change's own fields (operation, path, parameter/property names, types) — generic templates, no invented vendor facts. - [JavaScript SDK](https://www.npmjs.com/package/agentpulse-sdk): `npm install agentpulse-sdk` — zero-dependency client (global fetch only) wrapping this API with ETag caching and typed errors, for agents that prefer plain functions over MCP. Source: `sdk/` in the AgentPulse repo. - [All Monitored Services](http://agentpulse-43di.onrender.com/v1/services): Full catalog of tracked cloud providers, APIs, and registries. - [Live Incident Feed](http://agentpulse-43di.onrender.com/v1/incidents): Active incidents with severity ratings and mitigation hints. ## My Dependency Manifest & Affected Answers (Auth Required) - [Dependency Manifest](http://agentpulse-43di.onrender.com/v1/agents/me/dependencies): `PUT` replaces your stored manifest: `{"dependencies": [{"service": "stripe", "package": "stripe", "version": "14.5.0"}]}` (package/version optional; unknown slugs rejected with 400). `GET` returns it. Once set, preflight adds a compact `impact` verdict ({ affected, incident_ids, change_ids, reason }) per manifest service — affected means an active incident or a breaking change from the last 90 days touches your stack. ## Webhook Alerts (Auth Required) - [Webhooks](http://agentpulse-43di.onrender.com/v1/agents/me/webhooks): `POST` subscribes a URL to `incident.created`, `incident.resolved`, and `change.detected` events (optionally filtered by `services` slugs — set it to your manifest services to hear only about your stack; omitted = all services). The create response shows the signing `secret` ONCE. Every delivery is a POST of `{"id", "type", "created_at", "data"}` with headers `X-AgentPulse-Event`, `X-AgentPulse-Delivery`, and `X-AgentPulse-Signature: sha256=` — verify it before trusting a payload. `GET` lists your webhooks (no secrets), `DELETE /v1/agents/me/webhooks/{id}` removes one, `GET /v1/agents/me/webhooks/{id}/deliveries` shows the last 20 delivery attempts. URLs must be https (http allowed only for localhost). Deliveries retry up to 3 attempts with backoff. ## Stateless HTTP MCP Server - [Stateless MCP Endpoint](http://agentpulse-43di.onrender.com/mcp): Streamable Model Context Protocol endpoint supporting both JSON-RPC 2.0 and Mcp-Method / Mcp-Name header routing. Send `Authorization: Bearer ` to identify your agent. - `check_dependency_health`: Mid-debugging stack health preflight (manifest-aware when authenticated). - `lookup_breaking_changes`: Deprecation and schema diff analyzer. - `get_service_status`: Instant single-service health check. - `get_my_impact`: Affected verdicts for every service in your stored dependency manifest (auth required; api_key argument accepted as fallback). - `report_outage_signal`: Crowdsource early 5xx anomaly detection. ## Good Citizen Protocol - **ETag & 304 Not Modified**: Always pass `If-None-Match: ` in subsequent requests. A 304 response costs 0 tokens, grants +2 points to your Good Citizen Score, and **does not consume daily quota** — conditional requests are never blocked, even at your limit. Your Good Citizen Score is reputational — the free-tier quota is a flat 100 counted requests/day for every agent and does not expand; higher volume is a paid tier. ## Pricing - **Free** (`free_verified`): 100 counted requests/day, flat for every agent. Zero-CAPTCHA registration, no card. - **Pro**: 10,000 counted requests/day — $29/month. - **Team**: 50,000 counted requests/day — $99/month. - Paid plans are sold through Stripe Payment Links and **activate automatically on payment**: open your plan's link and append `?client_reference_id=` (the id from registration — or read `GET /v1/agents/me/plan`, whose per-plan `checkout_url` already carries it) so the payment maps to your API key. Pro checkout: https://buy.stripe.com/eVqcN5gxn4To9Sl8qO24000 — Team checkout: https://buy.stripe.com/6oUcN52Gx3PkaWpgXk24001. Keep the `client_reference_id` in the URL; without it the payment cannot be matched to your agent. Cancelling the subscription downgrades your plan back to Free automatically. Check your current tier, usage, and the full plan list any time at `GET /v1/agents/me/plan` (auth required; reading your plan does not consume quota). ## Quotas & Auth (Enforced) - **Daily quota**: free tier 100 counted requests/day, resetting at UTC midnight. Only identified (API-key) calls count; public unauthenticated reads (`/v1/status/*`, `/v1/services`, `/v1/incidents`, `/v1/changes`, discovery) stay open and uncounted. - **429 responses**: an exhausted quota returns `429` with `Retry-After` (seconds to reset) and a `rate_limit` body. Authenticated responses carry `X-RateLimit-Limit`, `X-RateLimit-Remaining`, and `X-RateLimit-Reset` (epoch seconds) headers. - **Writes require your API key**: `PUT /v1/agents/me/dependencies`, `POST /v1/telemetry/report`, `POST /v1/synthetic/probe`, and the MCP `report_outage_signal` tool all reject anonymous callers with 401. Registration itself stays zero-CAPTCHA and open.