Build an agent that calls tools to answer questions.
financial-assistant.prompt.mdx defines an agent with two tools: calculate and lookup_conversion_rate. The max_calls: 3 setting lets the agent make up to 3 tool calls before responding.
Tools are defined inline in the frontmatter with JSON Schema parameters — no separate registration needed.
agentmark run-prompt agentmark/financial-assistant.prompt.mdx
agentmark run-prompt agentmark/financial-assistant.prompt.mdx \
--props '{"question": "What is 15% tip on a $85 dinner bill?"}'Note: When running via the CLI, tool definitions are included in the formatted prompt but tool execution requires a tool registry in your
agentmark.client.ts. The CLI shows what tool calls the model wants to make. To handle tool calls end-to-end, use an adapter with a tool registry — see the docs.
- Tools are defined in the frontmatter — no separate file or registration needed
max_callscontrols how many tool invocations the agent can make in a loop- The adapter translates tools into the format your SDK expects (Vercel AI, Claude SDK, etc.)
- Tool parameters use standard JSON Schema — the same schema you'd write for an API