Feature

Your agent can read anything. It cannot tell a buyer from a bystander.

Whatever you already run, it can hand us text and get a verdict. Two REST calls, or MCP if you prefer.

Agents got good at reading in 2026. Grok Bot drives a browser and logs into whatever you grant it. OpenClaw plugs into anything with a plugin. Claude Code, Cursor and Windsurf speak MCP. A Python script with forty lines of feedparser reads a subreddit perfectly well.

None of them has an opinion about buying intent, and asking them to form one in a single pass gets you a confident answer exactly where confidence is least deserved. That is the gap, and it is the same gap regardless of which agent you picked.

SignalPipe takes the text your agent already has and returns a verdict, a role, and a drafted reply. We do not connect to your sources, do not hold your platform or model keys, and do not send anything on your behalf.

If your agent speaks MCP, there are 18 tools at the endpoint. If it does not, it is a REST call. Neither path is favoured, and there is nothing to rip out to try it.

How it works

1

Agent-agnostic by construction

We never fetch. Whatever reads the text is yours, which means the integration surface is the same for a browser agent, a plugin, an MCP client or a cron job.

2

MCP when you want it

18 tools at api.signalpipe.io/mcp. Works with Claude.ai, Claude Code, Cursor, Windsurf, or any MCP client, with your own operator key.

3

REST when you do not

Two calls. No plugin, no framework, no SDK. A script that already scrapes can add scoring without changing how it scrapes.

4

Batch for volume

25 posts per call with a per-post verdict and a spend summary, for agents working a queue rather than one message at a time.

5

Your credentials stay yours

Sending runs on your side from your own accounts. We hold no platform keys and no model keys.

See it in action

Join the waitlist. First 100 get Starter at 50% off for 3 months.

No credit card. No commitment.

Related