Connect your coding agent through MCP

Create a project credential

Open the intended project in Tideflow AI, then Settings → Connect your site builder. Create a token and copy it while it is displayed. Use the endpoint shown alongside it. A token selects one active project; you do not pass a project ID with every tool call.

Store the token in your coding agent’s private configuration or secret store. Do not put it in browser JavaScript, public documentation, or your repository. Rotating a token immediately revokes the previous one, so update connected clients after rotation.

Configure a Streamable HTTP connection

Add a remote MCP server in your client. Use the hosted URL below, select Streamable HTTP, and supply your token as a Bearer authorization header. Client configuration file formats differ; these are the connection values, not a universal client configuration file.

The MCP client handles initialization, protocol negotiation, and tool discovery. The endpoint is an MCP interface, not a general-purpose REST content API.

Endpoint: https://app.tideflow.ai/mcp
Transport: Streamable HTTP
Authorization: Bearer YOUR_MCP_TOKEN

Confirm the available tools

After connecting, ask your client to list Tideflow AI’s tools. Content delivery uses list_content, get_content, and ack_content. Author delivery uses list_authors, get_author, and ack_author. get_content_architecture supplies route context. get_analytics_setup and get_crawler_setup return project-specific installation instructions.

A useful first read is list_content. The JSON below is an MCP tools/call message sent through an initialized connection. It is not a command to paste into a shell or a standalone request that replaces MCP initialization.

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "list_content",
    "arguments": {
      "limit": 10
    }
  }
}

Give your agent a bounded task

Tell your agent which repository and website it is updating. Have it read the published snapshot, preserve the supplied route and metadata, validate the result, and acknowledge only work it successfully applied. Keep deployment within your team’s normal review and release process.

Use the connected Tideflow AI MCP server.
List pending published content for this project.
Fetch each snapshot and apply it at its supplied route.
Preserve its Markdown, author, hero image, SEO, and JSON-LD.
Validate the page using this repository’s checks.
Acknowledge only the exact revision successfully applied.
Leave failed items pending and report the failure.

Resolve connection problems

If authentication fails, check the endpoint, Bearer header, selected project, and whether the token was rotated or revoked. If your client only supports a different MCP transport, use one that supports this Streamable HTTP endpoint.

An empty content queue is not a failed connection. Pending published snapshots and deletion notices appear. Saving a draft does not publish it, and previously acknowledged revisions are no longer pending. Publish an approved article in Tideflow AI when you are ready to test the content handoff.