Integration
Integrations
Wire Brain into your stack — expose it to MCP clients, plug it into LangChain as memory, or delegate to it from OpenAI Assistants via function calls.
Brain ships a wire protocol, an HTTP edge, and hand-written SDKs — it does not ship a bundled MCP server or a first-party LangChain package. Each path below is a thin layer you wire over the SDK; the pages show the exact glue.
Pick a path
When to pick which
MCP tools
SDK / framework glue
Best for
Agentic clients (Claude Desktop, Cursor) where the LLM autonomously decides when to call Brain
Application code where you control the call sites
Setup effort
A small MCP server wrapping the SDK, plus one client config file
Library install + glue code
Auth
One key per MCP server instance
Per-request key, or one pooled key with per-request act_as
Context cost
Tool calls consume agent context
Zero context overhead — calls are explicit
Was this page helpful?
Backups
Back up the data volume — arena, WAL, and redb. Snapshot and restore mechanics, WAL-based point-in-time recovery, and the restore drill.
Expose Brain over MCP
Wrap the encode / recall / forget verbs as Model Context Protocol tools so Claude Desktop, Cursor, and other MCP clients can call Brain.