Live now

The social publishing API built for AI agents

Let your agent post to 10 networks. Works with Claude, OpenClaw, Hermes, Cursor, or anything that speaks MCP or can make an HTTP request.

Get an API key MCP setup REST API

One endpoint, either way

MCP if your agent speaks it, HTTP if it does not. Same auth, same checks, same three capabilities: see what is connected, publish or schedule, read back what happened.

MCP    https://attentionpilot.com/api/mcp
REST   https://attentionpilot.com/api/v1/accounts
       https://attentionpilot.com/api/v1/posts

Auth   Authorization: Bearer pilot_…

Connect your agent

Each of these is the same server with that client's config shape around it.

Claude

Claude Desktop, Claude Code, and anything else speaking MCP.

Set up Claude

OpenClaw

An open-source assistant that runs on your own machine and gains capabilities from skills.

Set up OpenClaw

Hermes

Nous Research's open-source agent, with persistent memory and a skills system that compounds over time.

Set up Hermes

Cursor, Windsurf, and everything else

{
  "mcpServers": {
    "attentionpilot": {
      "type": "http",
      "url": "https://attentionpilot.com/api/mcp",
      "headers": { "Authorization": "Bearer pilot_your_key_here" }
    }
  }
}

Or just a command

Every agent can run a shell command, whether or not it speaks MCP. The CLI has no dependencies, so it starts in about a second, and its exit code means something — a post where any network failed exits 1.

npx attentionpilot-cli login
npx attentionpilot-cli post "Shipped the thing." --to x,bluesky
npx attentionpilot-cli accounts --json

Not published to npm yet — that needs an account in Alex's name. The source is in cli/ and runs today with node cli/bin/pilot.mjs.

The skill

For agents that read skill files rather than call MCP, AttentionPilot ships one. It is a single Markdown file: the endpoints, the failure modes worth knowing, and four rules — the first being that the agent shows you the caption and the network list before anything goes out.

# In the AttentionPilot repo
skills/attentionpilot-social/SKILL.md

# Point your agent at it, and set
PILOT_API_KEY=pilot_…

Not on a skill registry yet — that needs an account in Alex's name. The file works today without one; a registry is distribution, not the product.

What stops it going wrong

Handing an autonomous process the ability to publish under your name deserves more thought than an API key and hope.

It cannot quietly half-publish

Every network is delivered on its own and the result names each one, so an agent is told exactly what landed and what did not — rather than a single success flag that hides four failures.

It is told no before it sends

The same pre-flight the composer runs happens on the API. A caption too long for X comes back as a 422 naming the network and the limit, before anything is scheduled.

It cannot spend what you have not got

Plan limits and per-network daily caps apply to agents exactly as they do to you. An agent in a loop hits a 402, not your reputation.

You can read what it did

Every delivery attempt is logged with its error text and shown per post in the app, so “why did that fail” has an answer that is not a guess.

Worth saying plainly

An agent posting on your behalf will eventually write something you would not have. Start it on one network, read what it produces for a week, and widen from there. AttentionPilot's brand voice narrows the range and the queue gives you time to look before anything sends — but nothing removes the need to read it.

Stop writing the same post four times.

Start free5 posts a month, no card.