MCP
Relay is a Model Context Protocol server. Point any MCP client at it with an API key and your agent gets four tools for making designs.
Add the server
In Claude Code:
claude mcp add --transport http graphtr \ https://www.graphtr.com/api/mcp \ --header "Authorization: Bearer gt_live_..."Or in a client that takes a config file:
{ "mcpServers": { "graphtr": { "type": "http", "url": "https://www.graphtr.com/api/mcp", "headers": { "Authorization": "Bearer gt_live_..." } } }}Tools
| Tool | Arguments | Returns |
|---|---|---|
| list_templates | category, kind, format, query, limit | Ids, names, dimensions, the slots each accepts and whether your plan can use it. |
| create_design | templateId or prompt, fill, preset, background | Renders a design and returns it as an image block the model can look at. |
| render_chart | type, rows, palette, showValues, background | Labels and values in, a designed chart out. Same engine as Charts. |
| get_usage | none | Plan, allowance used and when the period resets. |
The agent sees what it made
create_design returns a PNG as an image block rather than a link, so a vision capable model can look at the result and decide whether to adjust the copy and try again. The inline image is downscaled to keep it out of the way of the context window; the REST API is the full resolution path.
Errors
A failed tool call comes back as a successful response carrying isError, so the model reads the reason and can retry. Only authentication problems fail at the HTTP layer, with a 401.
Limits
Rendering counts against the same monthly allowance as the API. Listing templates and checking usage do not, so an agent can look around freely before it commits to making something.