Getting StartedDocs Search MCP

Docs Search MCP - Connect AI Tools to the Workflows Docs

Connect Claude, Cursor, or any MCP client to the Deepvue Workflows documentation. Search the API reference from your AI coding tools without leaving your editor.

Overview

These docs ship an MCP (Model Context Protocol) server, so external AI tools can search and retrieve this documentation directly. The server exposes a single read-only tool, search_documentation, which runs a hybrid search over the same indexed content this site renders.

That means Claude, Cursor, and any other MCP-compatible client can answer questions about the Workflows API with grounded, current documentation rather than from memory.

This is the read-only documentation search server. It is separate from the Deepvue MCP that runs live verifications, and it needs no credentials.

Endpoint

https://docs.deepvue.link/_mcp

The platform also serves the same server at its default hostname, https://decisioning.documentationai.com/_mcp. Prefer the custom domain above: it is the canonical one and it stays stable if the platform hostname changes.

Capabilities

FeatureDetails
Server nameDeepvue Workflows
Toolsearch_documentation
SearchHybrid, blending semantic and keyword matching
Search scopeEvery page, heading, API reference section, and code snippet on this site
AccessRead-only. No content modification
AuthNone required

Connect to the MCP server

Add the server to your project's .mcp.json:

{
  "mcpServers": {
    "deepvue-workflows-docs": {
      "type": "http",
      "url": "https://docs.deepvue.link/_mcp"
    }
  }
}

Or add it from the command line:

claude mcp add --transport http deepvue-workflows-docs https://docs.deepvue.link/_mcp

Tool reference

search_documentation

ParameterTypeRequiredDefaultNotes
querystringYesWhat to search for
limitnumberNo10Maximum results to return
semanticRationumberNo0.75Blend of semantic against keyword matching, from 0 to 1. Higher favours meaning, lower favours exact matches

Each result is a matched section, not a whole page.

FieldNotes
urlDirect link to the matched section
parent_urlThe page containing it. Fetch this when you need the full page
page_titleTitle of the parent page
section_titleTitle of the matched section
breadcrumbNavigation path to the section
contentSection content as plain text
llm_markdownThe same content as markdown, shaped for model consumption

The response also carries a count of results returned.

Because results are sections rather than pages, a broad question can match several sections of one page. Follow parent_url when you need surrounding context.

Example usage

Once connected, ask your AI tool questions like:

  • "How do I resolve a review on a Deepvue Workflows run?"
  • "What are the six run statuses, and which are terminal?"
  • "What is the difference between an abandoned run and an expired one?"
  • "Which run error codes are worth retrying?"
  • "What does the report content model return for a flagged step?"
  • "How do I configure a lifecycle webhook for a workflow?"

How it works

The server indexes public documentation only. Everything searchable through MCP is content already published on this site.

Next steps