Preview
What it is.
It takes you from an API with no connector to a working MCP server across four phases, recommending TypeScript with streamable HTTP and stateless JSON, and pointing you at the protocol spec to read before you write anything.
For a marketing team this is how the tool holding your pipeline data stops needing a weekly export. The quality bar is whether an agent can finish a task with it, so tool naming, concise descriptions, pagination and filtering all count, and error messages have to suggest a next step.
What you get.
- Four phases from research through implementation, with a stack recommendation
- Tool naming and description guidance, using consistent prefixes
- Context management rules, so tools return focused results instead of everything
- Error message guidance aimed at what an agent should try next
- Language guides for TypeScript and Python, plus a best practices reference
How to set it up.
Install the example-skills plugin from the anthropics/skills marketplace.
Decide between workflow tools and full endpoint coverage. The skill's advice when you are unsure is to prioritize coverage, because it gives the agent room to compose.
Read the protocol specification through the sitemap, fetching pages with a markdown suffix so they arrive readable.
Pick TypeScript unless you have a reason. It is the recommended language, with streamable HTTP and stateless JSON for remote servers.
Name tools with a consistent prefix and an action, such as a service name followed by the verb, so an agent can find the right one.
Add pagination and filtering from the start. A tool returning everything spends the agent's context on data it did not need.
Pricing Plans
Free. Apache 2.0 license.
Included in Anthropic's example-skills plugin. Checked against the repo in July 2026.
Use cases
Wrapping the billing API
Your billing tool has an API and no connector. Feed the skill the API documentation and work through its four phases, and a TypeScript server comes out that Claude can call directly.
Retiring the weekly export
Pipeline questions currently wait for a Monday export. Wrap the tool holding that data in a server, and from then on the numbers get pulled live in conversation.
Design pass on a rough server
You built a server and agents fumble it. Run the naming, description, pagination and error-message guidance over it, so each tool returns focused results and every failure names a next step.
Coverage or workflows, decided
You're unsure whether to build a few workflow tools or wrap the whole API. The skill's stated advice when in doubt is coverage, because it leaves the agent room to compose.
Best for
A CRM or billing tool with no connector
An API and no MCP server is the exact gap this closes, and most GTM data still sits behind one.
Replacing a weekly export
Once the server exists, the question gets asked in a sentence and the export step disappears.
Building an internal tool for the whole team
One server, and every skill on this page can query it, which is a better payoff than a one-off script.