Ask Claude, ChatGPT or your own app what happened. It reads your trace.

Base: https://pressfront.co/api/v1. Every call needs Authorization: Bearer pf_live_.... Answers are JSON (the brief is Markdown). The grey word is the same thing as an AI tool. Machine-readable: openapi.json.

A new trace takes a few minutes (POST /traces answers with etaSec). GET /traces/:id?wait=25 holds the answer until it finishes or 25 seconds pass, so repeat it instead of polling; the trace.done webhook works too. While it runs, /traces/:id shows soFar (accounts, critical, stage), and ?partial=true on /brief, /story, /accounts and /posts reads the first results, usually within about two minutes. They are for name traces; a story trace has its answer on /traces/:id/case and its brief on /traces/:id/brief.

Sides are read by a language model from each post's text, and every answer says so. A reading, not a finding: read the posts before acting on one.

Lists come in pages: pass the next value back as cursor (or before on /traces and /alerts) until next is null. Each post carries source: post (read on its own), saved (a saved copy, read with its account; traces from before Oct 24, 2026) or link (only the link was kept).

The MCP address is https://pressfront.co/api/mcp (streamable HTTP). It offers every endpoint as a tool, and tells the AI to quote posts rather than label people. Two ways in:

• Sign in with Pressfront (OAuth 2.1 with PKCE, dynamic client registration): claude.ai and any app that supports it. Discovery at /.well-known/oauth-authorization-server. A connected app shows in Settings as a connected app; Disconnect there cuts it off at once. Read only unless you tick the box on the Allow screen.
• Your key as Authorization: Bearer pf_live_...: everything else, below.

Claude Desktop: Settings, Developer, Edit Config, then add:

{ "mcpServers": { "pressfront": {
  "command": "npx",
  "args": ["-y", "mcp-remote", "https://pressfront.co/api/mcp", "--header", "Authorization:${PF_KEY}"],
  "env": { "PF_KEY": "Bearer pf_live_YOUR_KEY" }
} } }

Cursor: ~/.cursor/mcp.json

{ "mcpServers": { "pressfront": {
  "url": "https://pressfront.co/api/mcp",
  "headers": { "Authorization": "Bearer pf_live_YOUR_KEY" }
} } }

ChatGPT (a custom GPT): Actions, Import from URL, https://pressfront.co/api/v1/openapi.json. Authentication: API key, Auth type Bearer. Privacy policy: https://pressfront.co/privacy.html.

Anything else: generate a client from the OpenAPI file, or call the REST endpoints directly.

For your own https link we POST JSON with a Pressfront-Event header and a Pressfront-Signature header. Slack and Discord links get one plain line instead.

{
  "id": "evt_lq3k9x2a7f1c9e",
  "type": "alert.spike",
  "created_at": "2026-11-05T21:14:00.000Z",
  "data": {
    "name": "@yourname", "platform": "x",
    "count": 42, "usual": 4, "window_min": 15,
    "level": 1,
    "link": "https://pressfront.co/account.html?open=01J...#trace"
  }
}

We wait 8 seconds for a 2xx. If it fails we try twice more, after 1 and 5 minutes, with the same id, so you can skip one you already handled. If 3 alerts in a row never land, we pause the link and email you; email alerts keep coming. Fix it and press Send a test in Settings: a test that lands turns it back on.

The signature header is t=<unix time>,v1=<hex>, where the hex is HMAC-SHA256 of t + "." + raw body with your signing secret (Settings, under your link). Node:

const crypto = require("crypto");
function fromPressfront(rawBody, header, secret) {
  const parts = Object.fromEntries(header.split(",").map((p) => p.split("=")));
  const want = crypto.createHmac("sha256", secret).update(parts.t + "." + rawBody).digest("hex");
  const fresh = Math.abs(Date.now() / 1000 - Number(parts.t)) < 300;
  return fresh && crypto.timingSafeEqual(Buffer.from(want), Buffer.from(parts.v1 || ""));
}

60 calls a minute and 5,000 a day per key, 5 keys per account. An AI's tool call counts as one call; connecting and listing tools is free. Reading traces costs nothing extra. Starting them: 20 a day through the API, and a Trace plan's first 300 traces each month are full; after that they run as lighter traces until the month resets (GET /me shows how many full ones are left). Every answer carries X-RateLimit-Remaining. Over the limit you get a 429 with Retry-After in seconds.

Every error is {"error": "code", "message": "plain words"}.

• Keep your key secret. Anything done with it counts as done by you.
• No harassing, threatening, stalking, doxxing or targeting anyone with alerts, the API, an AI connected to it, or trace results. No building or selling a database of people. No reselling trace results as your own service.
• Posts in a trace belong to the people who wrote them and the platforms they were posted on. Follow those platforms’ rules for what you do with them.
• Alerts are an early warning, not a guarantee: we try hard, but we do not promise every alert arrives or every post is counted.
• We may pause or delete a key or link that breaks these rules. Keys stop working while a plan has ended.
• v1 stays v1: we give notice here before changing an endpoint in a way that breaks you. New fields can appear at any time; ignore the ones you don't use.

The full wording is in our Terms (section 7) and Privacy Policy (what we log and for how long).

More names, more calls, or something your team needs wired in: support@pressfront.co. Tell us what you are building.