Quickstart: shared context across AI tools

What you will do

Save a short project brief in one AI tool and read it from another. Webcraw is a shared knowledge store exposed through the Model Context Protocol (MCP), which lets your AI call tools to read and write that knowledge.

You explicitly ask to save and retrieve context. Connecting Webcraw does not import your chat history or automatically insert stored knowledge into every conversation.

1. Create your workspace

Create a free Webcraw account and verify your email. The organization name identifies your workspace; for a personal trial, use a name you recognize. The organization slug is generated from that name.

Use the same Webcraw account and workspace in both clients. You need an AI client that supports remote MCP connections and permission to add a connector. A managed account may require your administrator to enable it.

2. Connect two AI tools

Choose a client for its setup instructions, then repeat for your second tool. Both connections need access to the same Webcraw workspace.

https://mcp.webcraw.ai/http

3. Save in Claude, retrieve in Cursor

In Claude, ask:

Use Webcraw to create a ROOT knowledge node named "Atlas project context" with a description and this project brief: We use PostgreSQL, store timestamps in UTC, and require review before deployment. If it already exists, read it first and update it without removing other context. Show me the saved node and confirm the tool call succeeded.

A ROOT node is an entry point your other tools can find through Webcraw’s overview. Wait for a successful write result before switching tools.

Open a new conversation in Cursor and ask:

Use Webcraw to read "Atlas project context". Summarize the stored decisions before we start. If you cannot find it, tell me rather than guessing.

Check that Cursor called Webcraw and retrieved PostgreSQL, UTC timestamps, and review before deployment. Seeing those details in a fresh conversation confirms that this client can retrieve the stored context.

Continue with the full guide to update a decision and verify it back in Claude.

If something does not work

• No Webcraw tools: check the server URL, connection status, and whether the connector is enabled for this conversation. Reconnect if authorization has expired.
• Context not found: confirm the write succeeded and both tools use the same workspace. Ask the AI to call Webcraw’s overview, then read the exact saved node name.
• The AI replies without reading: explicitly ask it to use the Webcraw tool. Check the tool result rather than relying on a claim that it remembers.
• Outdated context: ask the tool to read the current node again. A chat may still contain an earlier copy until you retrieve the update.