Thanks to visit codestin.com
Credit goes to docs.context.dev

Skip to main content
The Context.dev skill helps a coding agent choose an endpoint, write a valid request, and handle the response. Install it once, then ask the agent to use it when building an integration. Want the agent to handle setup? Start with Agent Quickstart.
A skill does not connect to your account or make calls by itself. Use the MCP server for authenticated tools, or provide CONTEXT_DEV_API_KEY to a trusted local runtime when the agent needs to run code.

​
Install

Install into the current project when teammates and automation should use the same checked-in instructions:
Choose a global install when you want the skill available across projects:
The installer uses the site’s skill index and places or links the skill into supported agents’ directories. Add --agent codex, for example, to select a specific agent. Use npx skills add --help to inspect the current options. Use the site URL for the installer, not /skill.md. The file URL is for review or a manual installation.

​
Review before installing

The hosted source is public:
For a change-controlled environment, download and review the file, then commit the approved copy to your project instead of automatically following future hosted updates.

​
Manual project install

Agents that support the shared .agents/skills convention can read this layout:
Agent-specific global paths can change. Confirm the current path in your agent’s documentation before a manual global install.

​
Verify the skill

Start a fresh agent session in the project and ask:
Using the Context.dev skill, show the canonical HTTPS request to retrieve a brand from stripe.com. Do not run it.
A loaded skill should use POST /v1/brand/retrieve with this body:
If the agent invents a dedicated domain endpoint or uses a GET request, check that:
  • the installed filename is SKILL.md;
  • the skill directory is one your agent reads;
  • the session was restarted after installation;
  • another Context.dev instruction file is not shadowing it.

​
Run code with your key

The agent quickstart can provision and save a key through auth.md, after you complete the browser claim. For an existing key, expose it to the process through a secret manager, .envrc, or ignored environment file:
Do not paste the key into the skill, agent prompt, repository, or chat transcript. Keep tool approvals enabled for state-changing monitor operations.

​
Update the skill

The skill is intentionally concise and routes the agent to the API reference for full schemas. When updating a pinned copy:
  1. Review the diff.
  2. Confirm canonical Brand, Web, batch, and monitor routes.
  3. Re-run your integration tests against the updated SDK or HTTPS contract.
  4. Restart the agent session so it reloads the instructions.

​
Next steps

Agent quickstart

Turn an integration goal into a scoped task for your agent.

Connect MCP

Let your agent use live Context.dev tools alongside the skill.

Read the skill

Inspect the public instructions before installing or updating them.