Brainbox
Persistent, portable memory for AI assistants — a private per-tenant memory graph (memories, lessons, sessions, documents) reachable from any MCP client.
Community: Submitted by a user or imported; check the owner before granting accessOnlineNo sign-inGlobalFreeRead-only
What it can do
- Whoami: Show who you are signed in as, and how many memories your brain holds.
- Remember: Store a memory in your brain.
- Learn Lesson: Store a durable, reusable lesson — decays slower than a memory and is evidence-tracked.
What data it sees
Do you need an account
No: the server works without sign-in
Persistent, portable memory for AI assistants — a private per-tenant memory graph (memories, lessons, sessions, documents) reachable from any MCP client. Sign up at https://brainbox.thexi.dev.
How to connect
How to connect Brainbox to Claude
The server address to paste into Claude settings:
https://brainbox--cif-ai.run.tools
- Open Claude (claude.ai or the desktop app).
- Go to Settings → Connectors.
- Click Add custom connector.
- Paste the server address copied below into Remote MCP server URL and click Add.
- In a chat, click + → Connectors and switch the new connector on.
Custom connectors are available on Free, Pro, Max, Team and Enterprise plans (Free is limited to one). On Team and Enterprise an organization Owner adds the connector first under Organization settings → Connectors.
Authorization
After you click Add, a sign-in window for the service opens. Sign in with your own account and approve access. Claude never sees your password.
How to connect Brainbox to ChatGPT
The server address to paste into ChatGPT settings:
https://brainbox--cif-ai.run.tools
- Open ChatGPT in a browser (chatgpt.com). A Plus, Pro, Business, Enterprise or Edu plan is required.
- Turn on developer mode once: Settings → Apps → Advanced settings → Developer mode.
- Open Settings → Connectors and click Create.
- Fill in the form: Name (anything), Description (one line about what the service does), MCP server URL (copy it below).
- Under Authentication choose OAuth if the service requires sign-in, otherwise None. Click Create.
- In a new chat open + → Apps/Connectors and enable the connector.
OpenAI has renamed this section before (Connectors → Apps/Plugins). If the label differs, search settings for "developer mode". On Business/Enterprise workspaces an admin must allow custom connectors first.
Authorization
On first use ChatGPT opens the service's sign-in window. Sign in and approve access.
How to connect Brainbox to Cursor
The server address to paste into Cursor settings:
https://brainbox--cif-ai.run.tools
Fastest: click Open in Cursor below and confirm the prompt.
Manually:
- In Cursor open Settings → Cursor Settings → MCP and click Add new global MCP server.
- Paste the JSON copied below into
~/.cursor/mcp.json(per project:.cursor/mcp.jsonin the repo root). - Save the file. The server appears in the MCP list; authorize it there if asked.
Authorization
If the service needs sign-in, an authorize button appears next to the server in the MCP list.
Server tool list (23)
Raw names from tools/list. Only developers need these.
| whoami | Show who you are signed in as, and how many memories your brain holds. |
| remember | Store a memory in your brain. |
| learn_lesson | Store a durable, reusable lesson — decays slower than a memory and is evidence-tracked. |
| recall | THE memory search — ALWAYS use this for anything about the user, their work, projects, people or past conversations. Semantic search over memories and lessons (vector + keyword fused, decay-aware, salience-ranked, confidence-graded, evidence-tagged, with an honest abstention gate). For 'what's the latest / recent / catch me up' use `recent` instead — recency isn't semantic. Stored FILES/documents live behind `find_document`, not here. |
| recent | List your most recent memories and lessons, newest first. Use this for any recency/ordinal question — 'what's the latest', 'catch me up', 'anything new', 'recent updates'. For questions ABOUT a topic, use `recall` instead. |
| graph_query | Run a read-only Cypher query against your brain (power tool). |
| add_evidence | Attach evidence to a lesson — supports or refutes it. Evidence gates a lesson (facts outrank hypotheses): enough support → [CONFIRMED]; a strong refute → [DISPROVEN]. |
| apply_lesson | Record that a lesson was genuinely applied. Promotes its retention (FSRS: the harder it was to recall, the bigger the stability gain) and increments apply_count. |
| correct_lesson | Record that a lesson was wrong or needs revision. Increments correction_count and flags it for review; optionally replaces its content (re-embedded). |
| correct | Correct something the AI remembered wrong — works on memories, ideas AND lessons (for lessons it generalises correct_lesson). With new_content it rewrites the item (re-embedded; the superseded text is preserved in its lineage). Without, it flags the item as wrong for review. |
| forget | PERMANENTLY delete an item from your brain — the GDPR verb, for memories, ideas, lessons and documents. Everything derived from it cascades: similarity/co-activation/concept edges, evidence entries, lineage (:Mutation) records, recall receipts and session-log traces. NOT recoverable — for 'this is wrong' use correct instead. |
| pin | Pin a memory or lesson so recall reliably surfaces it — an explicit 'this matters' signal that outranks organic salience until unpinned. |
| unpin | Remove a pin — the item returns to organic salience ranking. |
| write_log | Audit what your AI wrote into your brain — every memory, idea, lesson and document with which client wrote it, how it authenticated, and when. Newest first; optionally filtered to one client. This backs the dashboard's audit view. |
| set_preference | Set a brain preference. Keys: capture_aggressiveness (off|sparse|balanced|dense — ambient auto-capture; off means the AI only writes when told), genealogist (on|off — idea-lineage recording), digest_cadence (off|daily|weekly). |
| capture_idea | Lightweight real-time idea capture — call DURING conversation when the user voices an idea, insight or decision. Lower friction than learn_lesson; stored as an embedded memory (kind: idea) so recall finds it. |
| store_document | Store a document/file (notes, reports, longer text) in your brain's file space (R2, tenant-isolated) and index it so find_document can retrieve it. For short conversational facts about the user, use `remember` instead. |
| find_document | Find or read stored FILES/documents only — NOT the user's memories (for anything about the user or past conversations use `recall`). Three modes: query (semantic search over titles/tags), file_id (read a specific file), folder (list a folder). |
| begin_session | Open a session with your brain (the wake of a dream cycle). Closes and distils any previous open session, returns a precomputed WakeBrief (what happened last time, open gaps, pending lesson proposals) and the brain-first protocol. Call at conversation start. |
| end_session | Close the session and let the brain dream: with a digest it distils shadow lesson proposals (never auto-written to your brain — review with list_proposals); without a digest only dedup and gap-logging happen. Call at conversation end WITH a digest. |
| list_proposals | List shadow lesson proposals the brain distilled from past sessions, for manual review. Nothing graduates into your real brain without resolve_proposal. |
| resolve_proposal | Manually review a shadow proposal: graduate it into a real Lesson (optionally edited) or reject it. This is the ONLY path from shadow to your real brain (MVP). |
| dream_metrics | Recall-as-dream health: sessions closed, digest rate, proposals/session, shadow precision inputs, replay hits, and the gate-escalation alarm. |