Docs · MCP

Use Vectorless from your AI apps

Connect Claude, Cursor, VS Code or any MCP client to your Vectorless workspace. Ask questions about your documents and get answers with every claim pinned to the exact sentence and page.

Server URL: https://mcp.vectorless.store/mcp

Connect an app

Claude Code

Run this, then type /mcp in Claude Code and sign in when the browser opens.

claude mcp add --transport http vectorless https://mcp.vectorless.store/mcp

Claude (web and desktop)

Settings → Connectors → Add custom connector. Name it Vectorless, paste the URL, and approve access.

https://mcp.vectorless.store/mcp

Cursor

Add to ~/.cursor/mcp.json. Cursor opens the sign-in page the first time it connects.

{
  "mcpServers": {
    "vectorless": {
      "url": "https://mcp.vectorless.store/mcp"
    }
  }
}

VS Code

Add to .vscode/mcp.json, start the server from the MCP view, and sign in.

{
  "servers": {
    "vectorless": {
      "type": "http",
      "url": "https://mcp.vectorless.store/mcp"
    }
  }
}

Scripts and servers

No browser to sign in with? Send a workspace API key as a bearer token instead.

curl -X POST https://mcp.vectorless.store/mcp \
  -H "Authorization: Bearer $VECTORLESS_API_KEY" \
  -H "Accept: application/json, text/event-stream" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Signing in

Apps sign in with OAuth: you approve them once, choose the workspace and what they may do, and can revoke them at any time under Connected apps in the dashboard. Apps that identify themselves by a metadata URL connect without registering first. Scripts use an API key from the dashboard instead; it carries the same permissions it was created with.

Tools

list_documentsList documentsdocuments:readread-only

List the documents in the workspace, newest first, with their status and size. Start here to find the document_id the other tools need.

get_documentGet a documentdocuments:readread-only

One document's details and processing status. After add_document, call this until status is ready (or failed) before asking questions.

get_structureGet a document's structuredocuments:readread-only

The document's table of contents as Vectorless verified it against the pages: every part, item and section with its page span. Use it to see what a long document covers, or to find a section to read in full.

read_sectionRead a sectiondocuments:readread-only

The full text of one section, with its title and page span. For reading a passage in context; to answer a question, ask_document is faster and cites its sources.

ask_documentAsk a documentqueryread-only

Answer a question from one document. Vectorless reads the document's structure, finds the passages that hold the answer, and returns the answer with every claim pinned to its exact sentence and page. Quote the citations when you relay the answer.

ask_workspaceAsk across documentsqueryread-only

Answer a question from several documents at once — all of them, or the ones you name. Use it when you don't know which document holds the answer, or to compare documents. Every claim is cited to its document and pages.

add_documentAdd a documentdocuments:writewrites

Add a document to the workspace from a public URL (PDF, Word, HTML, Markdown or text, up to 100 MB). Returns at once with status pending; processing takes seconds to a few minutes. Follow it with get_document.

delete_documentDelete a documentdocuments:writedestructive

Permanently delete a document and everything derived from it. This cannot be undone; confirm with the user first.

Errors

A failed tool call returns a typed error — NOT_FOUND, NOT_READY, QUOTA_EXCEEDED, RATE_LIMITED and so on — with whether it is worth retrying and when, so an agent can act on it instead of guessing.