MCP server

Connect Claude, ChatGPT or Cursor to Tatu

Tatu monitors how AI assistants — ChatGPT, Gemini, Perplexity, Claude — talk about your brand, and produces content to improve that visibility. Through the Model Context Protocol (MCP) any compatible assistant can read that data and work on your content, straight from the chat.

It is available on every plan, including the free one. You sign in with your Tatu account in the browser: there is no key to copy and nothing to install on your side.

Last updated: 2026-09-09

What you can do

  • Ask how your brand's visibility changed, which prompts mention it and which don't, and how you compare with competitors.
  • See which websites the AIs cite when answering your monitored prompts — and which ones just started showing up.
  • Run your monitored prompts right now and read the answers.
  • Turn a brief into content ideas, generate the piece, edit it with Tatu's editing AI and — if you grant that permission — publish it to your CMS.

Before you connect

  • A Tatu account (any plan) with a workspace that has at least one brand and one project.
  • The assistant always works inside the workspace of the account you sign in with. It cannot see other workspaces, and you never have to type ids.
  • Accounts without an active subscription can read everything; writing tools answer that the plan is inactive.

Connect Claude (web, desktop, mobile)

  1. In Claude, open Customize → Connectors and choose Add custom connector. (When Tatu appears in Claude's connector directory, you can add it from there in one click.)
  2. Name it Tatu and paste the server URL: https://project-honey-api.fly.dev/mcp
  3. Click Connect and sign in with your Tatu account. Claude shows the permissions it is asking for — only your identity — and you approve.
  4. Start a chat and ask, for example, how your brand did in the last 30 days. Claude runs read-only tools on its own and asks you before any change.

Claude Code (terminal):

claude mcp add --transport http tatu https://project-honey-api.fly.dev/mcp

Connect other clients

Cursor — add to your mcp.json:

{
  "mcpServers": {
    "tatu": {
      "url": "https://project-honey-api.fly.dev/mcp"
    }
  }
}

ChatGPT — Settings → Connectors → Advanced → Developer mode, then create a connector with the same URL. Any client that speaks MCP over Streamable HTTP with OAuth works the same way.

Example prompts

  • How did my brand's visibility in ChatGPT and Gemini change in the last 30 days?
  • Which websites do the AIs cite most when they answer my monitored prompts, and which ones are new this month?
  • Compare my share of voice with my competitors in the project [name].
  • Run my prompts about [topic] now and tell me which answers mention my brand.
  • Summarize this week's insights and tell me which ones need action.
  • Suggest three blog post ideas from this brief: [brief]. Save the best one and approve it.
  • Open the piece [title], ask Tatu's editor to make the intro answer the question in the first sentence, and apply the edits.

Tools

Each tool does one thing and says whether it reads or writes, so the assistant can run reads on its own and ask before changes.

Read — visibility and monitoring

ToolWhat it does
list_brandsBrands in your workspace.
list_projectsProjects — a brand with its prompts and competitors.
get_visibility_metricsMentions, citations, sentiment and rank for a project in a time range.
get_brand_timeseriesDaily series of one metric for a brand.
get_share_of_voiceShare of voice against the competitors.
list_promptsMonitored prompts and their latest results.
get_prompt_resultsThe answer each AI gave to one prompt.
list_citation_sourcesDomains the AIs cite, including the ones that just appeared.
list_insightsWeekly insights and recommendations.
list_competitorsCompetitors tracked in a project.
get_search_console_summaryGoogle Search Console summary, when connected.
get_executionStatus of an execution you started.

Read — content

ToolWhat it does
list_content_ideasContent ideas by status.
suggest_content_ideasThree idea suggestions from a brief. Nothing is saved; uses plan quota.
list_content_piecesPieces of a project.
get_content_pieceOne piece with its full body, in windows for long texts.
list_content_versionsVersion history of a piece.
list_wp_categoriesWordPress categories of the connected site.

Write — Claude asks before running these

ToolWhat it does
execute_promptsRun up to 20 prompts now. Uses plan quota; 20 per day through assistants.
create_content_ideaSave an idea (usually one of the suggestions).
approve_content_ideaApprove an idea so it can be generated.
generate_content_pieceGenerate the full piece. Asynchronous; uses the content quota.
update_content_pieceReplace the body — as a new version.
apply_content_editsExact find → replace edits — as a new version.
chat_with_content_editorOne turn with Tatu's editing AI. It proposes edits; they are applied only when you say so.
restore_content_versionBring back an older version — as a new version.

Publish — needs an extra permission

ToolWhat it does
publish_content_piecePublish to the connected CMS (WordPress, Ghost, Contentful, Framer, Wix or webhook). Only visible when the connection was granted the content:publish permission.

Permissions and safety

  • Sign-in uses OAuth through Tatu's identity provider. The assistant receives a token for your identity only; it is scoped to your workspace and can be revoked at any time by disconnecting the connector.
  • Read-only tools run without asking. Every writing tool is marked as such, so Claude asks you before running it.
  • Nothing is overwritten: every edit becomes a new version of the piece, and you can restore any earlier one — from the chat or from Tatu's editor, which shows where each version came from.
  • Publishing is the only action with an external effect. The publish tool is hidden unless the connection was explicitly granted the content:publish permission, and the assistant asks for confirmation before each publish.
  • Limits: 120 read calls per minute per user, 10 quota-consuming calls per minute, and a daily budget per workspace (1,000 requests on Free/Start, 10,000 on Growth, 50,000 on Scale). Prompt executions from assistants are capped at 20 per day and count toward the plan's monthly quota.

Data and privacy

  • The server only exposes data that already belongs to your workspace in Tatu: brands, projects, prompts and their results, insights, content. Your assistant sees what you would see in the app.
  • Tatu does not receive or store your conversation with the assistant. The exception is what you send to the editing AI (chat_with_content_editor): those messages are part of the piece's chat history, exactly as if you had typed them in Tatu's editor.
  • Tool calls are logged with identifiers and counts (which tool, which workspace, how long) — never with arguments or content — to enforce the limits above.
  • Everything else is covered by Tatu's Privacy Policy (https://www.usetatu.com/privacy) and Terms (https://www.usetatu.com/terms).

REST API

For scripts and integrations there is also a REST API on the same host (/v1), documented at https://project-honey-api.fly.dev/v1/docs. API keys are created in Settings → API & MCP and are available on the Growth and Scale plans.

Support

Questions, problems connecting or a client that does not work: suporte@usetatu.com.