How to share project context between Claude and Cursor

Carry a decision from planning to implementation

You settle on a database and deployment rules in Claude. When you open Cursor to implement the project, those decisions are still in the other conversation. This walkthrough puts a small project brief in Webcraw so both tools can read and update it.

The example project is called Atlas. You will save three decisions, retrieve them in a fresh Cursor conversation, change one, and read the update back in Claude. These are example prompts and expected checks, not a recorded customer session.

Before you start

You need a Webcraw account and Claude and Cursor connected to the same Webcraw workspace. Follow the public quickstart for the server URL, Cursor configuration, and Claude authorization steps. The shared-context example fits within the free tier; it does not use scheduled tasks.

1. Save a small, explicit brief in Claude

Enable the Webcraw connector in Claude and send:

Use Webcraw to save our Atlas project context. First check whether a knowledge node named "Atlas project context" exists. If it exists, read it and preserve unrelated content. Otherwise create it with the ROOT label so it is discoverable.

Save these decisions:
- Database: PostgreSQL.
- Timestamps: store in UTC.
- Deployment: require review before deployment.

Include a description: "Architecture decisions and working conventions for Atlas."
Show me the saved content and confirm the write succeeded.

Check the tool result for the stored node. A sentence saying “I’ll remember that” is not enough: the knowledge needs to be written to Webcraw. A named brief is also easier to find than an entire conversation.

2. Read the brief in a fresh Cursor conversation

Start a new Agent conversation in Cursor with Webcraw enabled. Do not paste the three decisions into it. Ask:

Read "Atlas project context" from Webcraw using its tools. If needed, use overview to locate the project, then read the node. Summarize its database, timestamp, and deployment decisions. If the context is missing or unavailable, say so.

Check the result: Cursor should retrieve PostgreSQL, UTC timestamps, and review before deployment from Webcraw. Inspect the read tool result to verify the source. You can now ask Cursor to use that brief while planning an implementation.

3. Update a decision in Cursor

Suppose the project now needs two reviewers for deployment. Ask Cursor:

Read the current "Atlas project context" from Webcraw. Update its deployment decision to require two reviewers before deployment. Preserve the database, timestamp convention, and any other stored context. Read the node back and show me the updated brief.

Updating the existing brief keeps the current decision in the same place. Check the read-back to make sure the other decisions remain intact.

4. Verify the change back in Claude

Read "Atlas project context" from Webcraw again, using a fresh tool call. What is our current deployment review requirement?

Expected result: the stored requirement is now two reviewers. Asking for a fresh read matters: an existing chat may still contain the earlier review requirement.

A habit for real projects

• At the start: ask the AI to read the relevant project brief from Webcraw.
• When a decision changes: ask it to update the existing brief and show the saved result.
• At the end: save a short handoff with completed work, open questions, and next steps.

Keep the brief focused on context you want other tools to access. Webcraw shares what you explicitly store and retrieve; it does not synchronize full chat histories or automatically refresh every conversation.

If the second tool cannot find your brief, use the quickstart checks to confirm the node exists and both connections use the same workspace.