---
name: knot-shared-memory
description: Read and save shared context with Knot so every AI tool the user works with knows the same decisions, preferences and project state, and gets the current version when something changes.
---

# Knot: shared memory for all your AIs

Knot holds what a user's AI tools have learned about them and their work: decisions, preferences, people, products, plans and what is in progress. ChatGPT, Claude, Cursor, Gemini and other tools read from and save to the same place. When a decision changes, Knot serves the new version and keeps the old one out of answers.

## When to use it

- Before answering, if the reply depends on background not established in this conversation (a name, a past decision, "where we landed", a project you have not seen), read from Knot instead of guessing.
- When something durable changes (a decision, a preference, a plan), save it once.
- When handing work to another AI tool, create a handoff so it can pick up where you left off.
- Skip Knot for self-contained turns.

## Connect

- MCP endpoint: `https://api.connectknot.com/mcp` (Streamable HTTP)
- Auth: OAuth 2.1 with dynamic client registration. Metadata: `https://api.connectknot.com/.well-known/oauth-protected-resource`
- Setup pages for each tool: https://connectknot.com/knot
- Server card: `https://api.connectknot.com/.well-known/mcp/server-card.json`

The user approves what each tool can read and save. Free accounts connect up to 2 AI tools; Knot Pro is unlimited.

## Main tools

| Tool | Use it to |
|---|---|
| `get_relevant_context` | Read the current facts and decisions relevant to a question or project. |
| `search_context` | Find a specific memory by keywords. |
| `get_context_recovery` | Catch up on what changed since this tool last read. |
| `propose_context` | Save one decision, preference or fact. Say what it replaces if it updates an older one. |
| `batch_propose_context` | Save several items, kept separate by project. |
| `create_handoff` / `check_pending_handoffs` | Pass work to another AI tool, or pick up work passed to you. |
| `report_active_state` / `get_active_state` | Share what you are working on right now, or see what other tools are doing. |

## How to talk about it

Tell the user in plain words what you read or saved, for example: "I saved your new pricing decision in Knot so your other AIs know too." Do not show internal IDs.

More: https://connectknot.com/whitepaper