Agent Authoring

The Diagrams smart tools are designed to be called from an agent. You describe the diagram you want in natural language; the toolset compiles it into a Graffiticode language and renders it through the shared embeddable runtime as SVG, viewable in any browser and embeddable in Figma or any documentation site.

How to connect an agent

The full integration walkthrough — MCP endpoint, free-plan access, discovery artifacts, one-call example, claim flow — lives at /agents. That page is the canonical "For AI Agents" reference for this toolset and is what you should point a new agent or a colleague at.

Supported agent platforms

Any MCP-compatible client can call the toolset; the platforms below have been tested and known to work out of the box.

• Claude Cowork — Anthropic's long-running agent surface. Add https://mcp.graffiticode.org/mcp as an MCP server — no credential required (leave the bearer field empty; the free plan needs no Authorization header).
• Claude Code — the CLI / IDE agent. Configure the MCP server in your .mcp.json or via claude mcp add.
• ChatGPT (with MCP) — register the endpoint in the ChatGPT connector or developer settings; ChatGPT will then be able to call create_item and the related tools.
• OpenClaw and other open MCP runtimes — same endpoint, no credential; see your runtime's docs for the exact registration step.

Diagram types

Once your agent is connected, browse the existing types: Concept Web, Venn diagram. Each page shows the kind of input the agent can describe and the kind of output it receives back.