Connect Memo to Claude with MCP

Memo is useful on its own, but it gets more useful the moment your AI assistant can reach into it. Instead of copying passages back and forth, you can let an assistant query your knowledge graph directly — search it, ask it cited questions, and read your sources — all from inside the tool you are already working in. The bridge that makes this possible is MCP, and connecting Memo takes a couple of minutes.

What MCP is, and why it matters

The Model Context Protocol (MCP) is an open standard for giving AI assistants access to external tools and data. An MCP server exposes a set of named tools; an MCP client — like Claude Code or another assistant — discovers those tools and can call them on your behalf during a conversation.

Why this matters is simple: it turns your assistant from something that only knows its training data into something that can act on your knowledge. Once Memo is connected, your assistant can pull real answers out of your own documents, with citations, rather than guessing or asking you to paste things in.

What Memo exposes over MCP

Memo runs an MCP server over HTTP. It exposes five tools:

• query — search your corpus by meaning and get back the passages that match.
• ask — ask a question and get an answer drawn from your documents, with citations.
• list_sources — list the documents and sources in your graph.
• get_source — fetch a specific source.
• get_graph — read the knowledge graph itself — nodes and the relations between them.

Between them, these cover the things you actually want an assistant to do with your knowledge: find, answer, enumerate, retrieve, and traverse.

Step 1 — create a Personal API Token

Requests to Memo's MCP endpoint are authenticated with a Bearer token. The recommended one is a Personal API Token, which you create in Settings. These tokens carry the prefix dgp_ so they are easy to recognise.

Create one and copy it now — you will paste it into your client configuration in the next step.

A session's Firebase token also works as a Bearer credential, but a Personal API Token is the right choice for anything you want to keep connected.

Step 2 — add Memo to your MCP client

Memo's MCP endpoint lives at the API base host, on the /mcp path. With Claude Code, adding it is a single command — supply the transport, a name for the server, the endpoint URL, and your token as an Authorization header:

claude mcp add --transport http memo https://docgraph-api-b7e2oxlrrq-km.a.run.app/mcp --header "Authorization: Bearer dgp_your_token"

Swap dgp_your_token for the token you created in Settings. That registers Memo as an HTTP MCP server called memo, and its five tools become available in your next session.

Clients that only speak stdio

Some MCP clients connect over stdio rather than HTTP. You can still use Memo from those — bridge the HTTP endpoint to stdio with mcp-remote:

npx mcp-remote https://docgraph-api-b7e2oxlrrq-km.a.run.app/mcp --header "Authorization: Bearer dgp_your_token"

Point your stdio client at that command and it will reach Memo through the bridge, with the same five tools available.

What you can now ask

With Memo connected, your assistant can work directly against everything you have filed. The questions that used to mean opening a dozen documents become single requests:

• "Search my corpus for anything about supply-chain risk and summarise what you find."
• "Ask Memo what this organisation's stated position on emissions is, and cite the sources."
• "List my sources from the last upload, then pull the full text of the second one."
• "Walk the graph out from this concept and tell me which entities it connects to."

Because ask returns cited answers, you can follow any claim straight back to the document it came from — which is what makes an assistant you can actually trust over your own knowledge.

Going further

This covers connecting and asking. For the full tool reference, request and response shapes, and more detail on authentication, see the MCP guide and the HTTP API reference. If you would rather work from the command line, the CLI is the docgraph-mcp Go binary, built from source — see the CLI guide.