xhumanSign in →

Connect via MCP

Bring your mind into Claude, ChatGPT, and every other AI client.

The xhuman MCP connector exposes your second brain to any MCP-compatible AI agent — search, read, ask, traverse the graph, recall memories, and, when you authorize it, add to and edit what's there. OAuth-protected, scoped to your own data, free.

Try asking

Recall what you've thought about a topic

“Search my mind for everything I've written about retrieval-augmented generation, then summarize the key tensions in my thinking.”

Get a citation-grounded answer

“Ask my mind: how have my views on remote work evolved over the last year? Cite the articles that support each shift.”

Find adjacent ideas you forgot you had

“Show me my knowledge graph filtered to the 'concept' type with authority above 0.7. Which concepts cluster together that I haven't explicitly linked?”

Trace a memory from a chat

“Search my memories for what I said about the new launch plan in last week's conversations. List the three most relevant.”

Build on a specific article

“Read the article 'second-brain-thesis' and use it to draft a 200-word LinkedIn post in my voice.”

How to connect

Claude.ai

  1. Open Settings → Connectors → Add custom connector
  2. Paste https://xhuman.com/api/mcp
  3. Click Connect. Sign in to xhuman in the popup. Authorize.
  4. Open a new chat. The connector's tools are now available.

ChatGPT

  1. Settings → Connectors → Advanced → enable Developer mode
  2. Apps & Connectors → Add MCP connector
  3. Paste https://xhuman.com/api/mcp
  4. OAuth flow signs you into xhuman; ChatGPT discovers the tools automatically.

Claude Desktop, Cursor, Claude Code, OpenClaw

Sign in at /mcp and use the platform-specific install panel. The page generates a personal token, formats the right config block, and copies it to your clipboard.

What it can do

Read your mind · 8 tools

  • search_mindHybrid search across your wiki articles. Returns metadata only — call read_article to drill in.
  • read_articleFetch a full article body by slug. Use after search_mind.
  • ask_mindRetrieval-augmented Q&A with inline [slug] citations. Runs a model pass, so it costs more than a search.
  • timelineYour chronological life timeline — events, eras, categories.
  • list_indexMetadata for every article in your wiki. Cheap overview.
  • get_graphYour wikilink knowledge graph as nodes + links.
  • search_memoriesSemantic search over atomic memories (deduped facts from chats and voice).
  • list_memoriesChronological slice of your memories, newest first.

Change your mind · 7 tools

  • ingest_sourceAdd text or a URL to your mind. A URL is fetched and converted, then compiled into articles.
  • record_factRecord a fact as an atomic memory. Deduplicated and merged against what you already have.
  • update_articleReplace the body of one wiki article, then re-index it for search.
  • delete_articleDelete a wiki article by slug. Permanent — it cannot be recovered.Removes data
  • update_memoryReplace the text of one memory and re-index it for search.
  • delete_memoryDelete a memory. It stops appearing in search and retrieval immediately.Removes data
  • correct_claimFix something wrong: the old claim is revoked and the correction takes its place.

The reading tools only return what is already there. The writing tools change your mind's contents — they add sources, record and rewrite memories, update and correct articles, and delete articles or memories. A deleted article cannot be recovered. Every tool declares to your AI client whether it writes, and the two deletions are declared destructive, so a client can decide when to ask you first; whether it asks is up to that client, and this server adds no approval step of its own. Nothing here sends email or messages or posts anywhere. The one outbound request is ingest_source fetching a URL you gave it.

Trust & safety

  • OAuth 2.1 with PKCE. No tokens to copy and paste — Claude.ai and ChatGPT discover and authorize automatically.
  • Audience binding. Tokens default to your private (`self`) audience. Mint a public-only token from /mcp to give an agent access only to claims you've marked public. Audience binding narrows what the reading tools return; it does not restrict the writing tools.
  • What an authorization grants. Connecting from Claude.ai or ChatGPT, or minting a personal token at /mcp, issues the mcp scope, which unlocks every tool above — the writes and the deletions included. The consent screen says so before you approve. A client that asks for read instead gets the reading tools only, and the server refuses write calls on that token.
  • Revoke anytime. From /mcp or inside the AI client’s connector settings.
  • Read the policy: MCP connector privacy policy

Support

Email support@xhuman.com with the AI client name, the prompt that failed, and any error message you saw. We'll respond within one business day.