Build with Gavana

Gavana works as a visual workspace for people and as a precise canvas contract for agents. Use the path that matches what you want to do; every path returns real canvas handles and review links, so you can always check the result on the board.

Choose a path

You want to…	Best path	Authentication	What it can do
Work with Gavana from ChatGPT or a remote MCP client	Hosted MCP	Browser OAuth	Inspect, save images, create/run workflows, generate image or video work after explicit approval
Let Codex or Claude work locally	Local MCP	Agent Access token or local CLI profile	Broad canvas, asset, Recipe, Image Action, image, video, and Run tools
Automate from a terminal	CLI	Browser OAuth or Agent Access token	Scriptable JSON output and the same stable Canvas API operations
Build your own service	Canvas API	Agent Access token	Stable OpenAPI contract, revision-safe writes, callbacks, and durable handles

For inspection only, use the hosted read-only endpoint: https://app.gavana.ai/mcp/readonly. It exposes only read operations and cannot save, generate, cancel, or alter a canvas.

A safe operating contract

This is the same guidance for a person using an agent and for an agent reading this page:

Inspect before deciding. List canvases, read the exact canvas, then use returned canvas:, node:, and asset: handles—never guessed IDs.
Ask before side effects. Saving an image writes a durable node. Workflow, image, and video generation can use a connected provider and incur cost. Never start or automatically retry paid work without explicit current-turn approval.
Keep writes safe. Reuse a caller-stable 8–200 character idempotency key for the same intended request. For raw canvas writes, read the latest baseRevision first.
Finish the handoff. Poll the matching Run or video status tool while work is queued or running, then return the durable node, asset, job/run handle, or review link.
Protect credentials. Keep OAuth and Agent Access tokens out of chat, URLs, screenshots, repositories, and logs. Share X-Request-ID with support instead.

Start here

Machine-readable references

This help center publishes an LLM index and complete help-center context alongside its normal pages. For exact API operations and schemas, use Gavana’s generated OpenAPI 3.1 document rather than inferring an endpoint or parameter.