Coding agents

Point Claude Code, Cursor, Codex or any coding agent at coss.com, and it adds email, SMS and webhooks to your project.

Give your agent one prompt:

Add coss.com to this project: read https://coss.com/llms.txt, get a development key with `npx coss-cli init`, and build and test with it.

From there the agent reads what it needs from the docs, asks you to approve a development key in your browser, installs the SDK and tests its code with that key. Nothing it sends is delivered until you switch to a production key yourself.

What agents read

  • /llms.txt starts with a short brief (how to get a key, how the SDK reports errors, how to test webhooks), then lists every docs page.
  • Every docs page is Markdown at its URL plus .md: /docs/email/sending.md. The HTML pages point to it too (<link rel="alternate" type="text/markdown">), and each has a Copy Markdown button for pasting it into a chat yourself.
  • API reference pages list each operation's method and URL, parameters, request body, responses and errors, with examples for the TypeScript SDK, HTTP and the other SDKs: /docs/api-reference/emails/send-email.md. The OpenAPI 3.1 document describes the same API.
  • /llms-full.txt is every guide in one file, with a short entry per API operation. It's large: agents usually do better fetching the pages they need.

Development keys

npx coss-cli init, run in the project, gets the agent a key without you copying one around: it opens the dashboard, you check the code and approve a key for that project (development environments only), and the CLI writes it to .env.local (Next.js) or .env as COSS_API_KEY and installs the coss package. Where no browser can open (over SSH, in a container or an agent's cloud sandbox), it prints a link instead and waits for the code that page shows, typed into the terminal: run it yourself there.

Development keys work with every API and never deliver anything; emails and SMS get simulated outcomes, events and webhooks (see sandbox behaviour). So the agent can run what it writes, and coss listen forwards the resulting webhooks to its local handler.

npx coss-cli init --agents also adds a short coss.com section to the project's AGENTS.md (or CLAUDE.md), so later sessions know where the docs are and how to test without reading this again.

Claude Code plugin

The coss.com plugin gives Claude Code a skill for building with coss.com (loaded when a task involves it) and the MCP server:

/plugin install coss --marketplace cosscom/coss

On Claude Code before 2.1.275, add the marketplace first: /plugin marketplace add cosscom/coss, then /plugin install coss@coss.

It asks for an API key for the MCP server. Use a development key (coss_test_…), or leave it empty: the server's docs tools work without one.

Cursor, Codex and other agents

The same skill installs into Cursor, Codex, GitHub Copilot, Windsurf and other agents with the skills CLI:

npx skills add cosscom/coss

Add the MCP server as well to let the agent search these docs, look up emails and SMS, and explain why one wasn't delivered.