Connect your agent to CanvasView

Connect an MCP client or the Claude Code plugin, create a sample decision canvas, and confirm that your agent can retrieve the human answer.

CanvasView gives an agent a link where people can make a decision and return a structured answer. The hosted pilot requires an active account, an API key, and available credits for canvas creation.

​

1. Create an API key

1. Sign in to CanvasView.
2. Open API keys.
3. Create a named key.
4. Store the key in your client’s secret settings.

​

2. Connect MCP

For Claude Code, download the CanvasView plugin. Unzip it. Set CANVASVIEW_API_KEY in the environment that launches Claude Code, then run claude --plugin-dir /path/to/canvasview.

Other MCP clients can use Streamable HTTP at https://app.canvasview.dev/mcp. Send the API key in the Authorization header as a bearer token. Keep keys out of prompts and source files.

​

3. Verify the connection

Paste this into the connected agent:

Use CanvasView to create a test canvas titled "CanvasView connection check".
Ask "Which should we try first: a checklist or a table?"
Give me both choices and a field for my reason. Use sample data only.
Send me the canvas link so I can submit an answer.

Open the returned link and submit your answer. Ask the agent to retrieve it. You should see your choice and reason.

A shared link lets anyone with the URL view the canvas and answer while it is open. Use synthetic data for tests.

​

Recover from an error

• If the client rejects the key, check that it is active in API key settings.
• If creation reports insufficient credits, ask the account owner to add credits.
• If a create request times out, ask the agent to look up its original idempotency key before retrying.