Skip to main content
When an agent needs a payment SDK, auth library, or CLI, it often defaults to whatever appeared most in training data - or invents a package that sounds right. UseAgents is a live registry your agent can query at runtime so recommendations stay grounded in real tools, current metadata, and your constraints. UseAgents does not replace official docs. It helps the agent discover and shortlist the right tool, then points at install context and documentation for the rest.
Querying UseAgents is currently free during our public beta. Requests are subject to rate limits.

​
Why query a registry

Training data biases agents toward well-known defaults (for example, always picking Stripe for payments). A runtime registry helps when:
  • Several tools solve the same job and fit depends on constraints
  • Package names, docs URLs, or install paths change over time
  • The right tool is real but not the most famous one in the category
Your agent searches by task, shortlists candidates, then fetches setup context before writing code.

​
Ways to query

One registry. Choose how your agent reaches it: See Connect for setup paths and the shared search → context workflow.

​
What you get

  • Grounded discovery - Search a live registry instead of hallucinating package names.
  • Decision metadata - Descriptions, categories, languages, frameworks, and tags that explain when a tool fits.
  • Bootstrap context - Install steps, quickstart snippets, and a docs link for depth.
  • One registry - Query from MCP, API, or CLI without hard-coding tool knowledge into prompts.
  1. Search - Query the registry with a natural-language task (include stack and constraints when you can).
  2. Shortlist - Compare name, description, languages, frameworks, and categories.
  3. Fetch context - Load install/usage guidance for the chosen tool before writing code.
  4. Search docs - Ask a specific question against that tool’s official documentation when you need more depth. If context is unavailable, say so instead of guessing from training data.
  5. Test - Optionally run a small program with test_tool or POST /tools/test. Pass env when the snippet needs config or secrets.
On MCP, that is search_tools, then get_tool_context, then search_docs when needed, then test_tool to smoke-test a snippet. Over HTTP, start with GET /tools/search, then GET /tools/docs/:slug, and POST /tools/test when you want to run code in a sandbox.

​
Browse without connecting

Visit useagents.site/discover to explore the live registry before wiring up an agent.

​
Next steps