Word of Agent
Word of Agent checks whether search engines and AI assistants can find, read and recommend a website, and says what to change. Connected to your assistant, it gives the assistant the measurements and the work: your assistant fixes, writes and drafts; you approve anything that goes out in your name; Word of Agent checks the result.
This repository holds what assistants and directories need to use the hosted service: the MCP server's registry entry, the plugin manifest and the agent skill with its playbooks. The service itself is not open source.
Connect it
The server is a remote MCP server with OAuth:
https://app.wordofagent.app/mcp
You sign in to Word of Agent and choose which account the assistant may use. A free account is enough to start: sign up at wordofagent.app with your email and add your site.
- Claude: Settings, Connectors, Add custom connector, paste the address, then Connect.
- ChatGPT: Settings, Apps and Connectors, Create (turn on developer mode under Advanced if you do not see it), and use the same address.
- Claude Code, Codex and other agents: add it as a remote MCP server with that address; they open the sign-in
in your browser. For example,
claude mcp add --transport http wordofagent https://app.wordofagent.app/mcp.
What your assistant can do with it
- Measure. Audit a site: crawler access, indexing in Google and Bing, rendering, structured data, the facts the site states, listings and reviews.
- Ask the customers' questions. Get the questions, answer them the way it would answer anyone, and send the answers; read them back for who is named instead, what is said wrong, and why.
- Fix. Work through the to do list most costly first, with the exact fix for each finding, then have Word of Agent check it again.
- Get the business heard. A main channel chosen from evidence, pages to write, sources to get mentioned on, local listings, reviews, rivals compared and listening.
- Draft for you to approve. Outreach, replies, review requests and pitches are drafts: nothing is sent, posted or published until you approve it in Word of Agent.
What it never does
- It calls no AI model: your own assistant does the thinking.
- It sends nothing in your name. An assistant cannot approve its own drafts or plans: only a person signed in to Word of Agent can.
- It does not scrape personal data or buy lists.
In this repository
| Path | What it is |
|---|---|
server.json | The entry in the MCP Registry (app.wordofagent/wordofagent) |
plugin.json, mcp.json | The plugin manifest and its MCP server |
skills/wordofagent | The skill for agents, with a playbook for each kind of work and a guide per site platform |
The skill and playbooks are also served by the connector itself (get_playbook).
Data and support
What we collect and why is in the privacy policy; the terms cover using the service. More at wordofagent.app/connector. Questions or problems: support@wordofagent.app.