Agentic Mermaid — diagram workflow

An agent-agnostic typed editing surface for Mermaid. New diagrams can be authored as Mermaid source and verified/rendered directly. Existing modeled diagrams can be parsed to a ValidDiagram, mutated with typed ops, verified structurally (not as subjective visual scoring), and serialized back to canonical source. Agentic Mermaid outputs SVG, PNG, ASCII, Unicode, and JSON layout; layout is deterministic — verified cross-process, no layout seed. Styled looks (style render option: name | spec | stack, e.g. ['hand-drawn', 'dracula']) accept an ink seed that re-rolls wobble without ever moving layout; see docs/style-authoring.md.

Pick a channel

• agentic-mermaid-mcp connected → Code Mode (references/code-mode.md). Multi-step edits in one round-trip.
• Can run JS/TS with imports → library (agentic-mermaid/agent). Same SDK.
• Shell only → CLI (references/cli.md).
• No local install, network only → hosted MCP at https://agentic-mermaid.dev/mcp (stateless Streamable HTTP JSON-RPC; execute, describe_sdk, render_svg, render_ascii, render_png, verify, describe, mutate, and build tools — 64 KB input cap). Read references/hosted-mcp.md before composing a hosted request. Use the named direct tool and its exact argument object; do not substitute library APIs or local Code Mode for a hosted direct-tool task.

For hosted MCP, choose the least-powerful tool that completes the task: