AI access · Model Context Protocol

Use BrainKB from your AI assistant

BrainKB exposes an MCP server so an assistant can operate the knowledge graph on your behalf: ingest data, search, explore provenance, and manage spaces, using your own account and permissions. Two ways to use it — point any MCP client (Claude, etc.) at the server directly, or use the BrainKB skill, which runs on the same MCP and adds guided, step-by-step workflows. Every action is access-controlled server-side exactly as in this web app. For example, just ask: “Ingest my TTL into the hmba space, then show what changed.”

Connect

Add the hosted MCP endpoint to your client (e.g. Claude Code):

claude mcp add --transport http \
  brainkb https://mcp.brainkb.org/mcp

Registry id org.brainkb/brainkb, transport streamable-http. Prefer guided workflows? The BrainKB skill wraps these same tools with step-by-step guidance in Claude Code.

Authenticate

Sign in once, mint a Personal Access Token, and send it with each request:

Authorization: Bearer brainkb_pat_…

In the skill: brainkb_create_token() after a one-time Globus login, then set it as BRAINKB_TOKEN. Tokens are revocable and expire on inactivity.

What the assistant can do
Spaces & graphs
Create private/team workspaces, register named graphs, manage members and access.
Ingest
Load RDF (Turtle / N-Triples / JSON-LD) — raw text or files up to ~5 GB — as background jobs.
Read & search
Access-filtered full-text search, read a space's RDF, list registered graphs.
Provenance (PROV-O)
Per-job provenance, a graph's change history, and the exact triples each ingest added.
Personal Access Tokens
Mint a browser-free token to authenticate the assistant, then revoke it any time.
Admin (if authorized)
Manage users, roles, capabilities, and per-space access rules.
The assistant acts as you — it can only see and change what your roles and space membership permit, and a 403 means a permission is missing, not a broken token. Full tool reference and examples: docs.brainkb.org.