Skip to main content
Three things let an AI agent discover Valendata and use it without help. Each one points to the pages it needs, so it never goes stale.

llms.txt

Agents that read llms.txt find Valendata in two places: Any docs page also reads as Markdown: add .md to its URL.

The Valendata Agent Skill

skills/valendata/SKILL.md is an Agent Skill. It loads when a task needs data from a website, and teaches the agent to:
  1. Find a skill for the task with GET /v1/marketplace/search?q= (MCP tool: search_skills). No key is needed for the public marketplace. Add kind=workflow to find a workflow (MCP tool: search_workflows).
  2. Read its inputs, sample rows, and price, in the search result or with Get a skill or Get a workflow.
  3. Clone one that is not yours with Clone a skill or Clone a workflow (MCP tools: clone_skill, clone_workflow). A public skill never runs directly: the API answers 403 clone_required.
  4. Run the copy and wait, or start it in the background and poll. See Run a skill.
  5. Handle 402, 429, and needs_user. See Errors and limits.
  6. Create a skill when none fits. See Create a skill.
It uses the MCP tools when they are connected, and the API with your key when they are not.

The Claude Code plugin

The agent kit is also a Claude Code plugin. It adds the skill above and the Valendata MCP server, signed in with an API key.
  1. Make an API key with only the scopes the agent needs. Add workflows:read and workflows:invoke if it should use workflows, and skills:write if it should create skills.
  2. Set it in your shell before you start Claude Code:
  3. In Claude Code, add the kit as a marketplace and install the plugin:
  4. Run /mcp to check that valendata is connected.
To sign in with OAuth instead of a key, or to connect another assistant, see Connect your AI assistant.