Your coding agent can read the file now

The most common way a design reaches a coding agent today is a screenshot pasted into a chat. The agent then guesses at spacing, invents a colour that is nearly right, and rebuilds a component you already have. We have shipped the alternative: a server the agent connects to, which hands it the design itself.

What it can fetch

Four tools, deliberately few. One identifies who is connected and what they are allowed to read. One returns a sparse outline of a document — ids, names, types, parents, bounds — for finding your way around. One returns the full structured context rooted at a node: the tree, the variables bound in it, the components it uses, the assets, in the shape of the framework you name. And one returns just the variable and style definitions, when that is all you need.

{  "mcpServers": {    "lindi": {      "url": "https://lindi.ai/api/mcp",      "headers": { "Authorization": "Bearer lmcp_…" }    }  }}# tools: whoami · get_metadata · get_design_context · get_variable_defs

The token is yours

Every connection uses a token you create yourself, in Dev Mode, for one file. The list shows when each one was created and last used, and any of them can be revoked — at which point the connection is gone the same moment. The server is read-only: an agent can read the design; it cannot change it.

Nobody should have to describe a screen to an agent twice. The file already knows what it is.

This is available now. If your engineers already run an agent, point it at a file and watch it stop guessing.