AI

Every SendMock feature is an operation that the UI, the REST API, the CLI and the MCP server all share. So anything you can click, an AI coding agent (Claude Code, Codex, Cursor, Gemini CLI, VS Code agents and others) can do too, with the same checks: create a gateway, compose a flow, write a template, add a rule, read an OTP, replay a delivery.

What an agent can do for youCopy Anchor Link

• Connect your app: find the mail and SMS code in your repository, point it at SendMock, send a test.
• Test a flow end to end: trigger your app's sign-up, wait for the SMS, read the code, finish the sign-up.
• Build what SendMock does not ship: a payment gateway from its docs, a checkout flow, a webhook template for a provider, outcome rules, inbound routes, catcher responses. See Build it with AI.
• Debug: read the latest failed delivery, explain the error, fix your handler, replay.
• Keep it: export the result to the seed file, or write a scenario that runs in CI.

Set it upCopy Anchor Link

SendMock not installed yet? Your agent can install it too: paste the prompt from Install with AI.

Pick your agent, then run its command:

sendmock ai install --target claude-code

What the command does:

• Registers SendMock's MCP server with your agent.
• Installs the agent skill, or an instructions section for agents without skills.
• Writes them in your project for Claude Code, Cursor and VS Code, and in your user settings for Codex and Gemini CLI.
• Shows what it will change and asks first. Add --yes to write without asking.
• Without --target, it prints every agent's install command and MCP configuration, for any other agent.

With npx or bunx (nothing installed)Copy Anchor Link

If you run SendMock with npx or bunx, there is no sendmock on your PATH, so the agent starts SendMock through npx or bunx instead.

1. Set up your agent. Pick your agent in the tabs.

   With npx (needs Node.js):

   Claude Code

   Codex

   Cursor

   Gemini CLI

   VS Code

   Copy Text

   npx sendmock ai install --target claude-code

   With bunx (needs Bun):

   Claude Code

   Codex

   Cursor

   Gemini CLI

   VS Code

   Copy Text

   bunx sendmock ai install --target claude-code

   This registers the MCP server and installs the agent skill, the same as the installed command does. The configuration it writes starts SendMock with npx -y sendmock mcp, so your agent needs Node.js even when you ran bunx. If you only have Bun, register the MCP server yourself as in step 2. From an installed SendMock, sendmock ai install --npx writes the same npx configuration.

2. Or register the MCP server yourself. Pick your agent; each example starts SendMock with npx:

   Claude Code

   Codex

   Cursor

   Gemini CLI

   VS Code

   Copy Text

   claude mcp add sendmock -- npx -y sendmock mcp

   With Bun instead of Node.js, use the command bunx with the arguments sendmock mcp. In Claude Code:

   Copy Text

   claude mcp add sendmock -- bunx sendmock mcp

3. On Windows, if your agent cannot start npx or bunx directly, put cmd /c in front: cmd /c npx -y sendmock mcp.

Settings > Connection > AI agents shows each agent's commands with the installed binary or npx: choose npx (no install) under SendMock runs as.

Claude Code pluginCopy Anchor Link

In Claude Code you can install the SendMock plugin instead.

• It carries the agent skill and registers the MCP server.
• Claude Code updates it with each SendMock release.
• Its MCP server runs sendmock mcp, so SendMock itself must be installed and on your PATH.

1. Add SendMock's plugin marketplace:

   Copy Text

   /plugin marketplace add shibbirweb/sendmock

2. Install the plugin:

   Copy Text

   /plugin install sendmock@sendmock

Any other agentCopy Anchor Link

Register SendMock's MCP server in your agent yourself.

1. Add an MCP server that runs this command (stdio):

   Copy Text

   sendmock mcp

2. Make sure SendMock is running. The MCP server talks to the SendMock at http://127.0.0.1:8025.

   • Another address: add --server <url>.
   • No SendMock running: add --embedded, and the MCP server starts SendMock itself. It stops when your agent session ends, and takes its settings from the SENDMOCK_* variables and sendmock.toml.

Other ways to connect:

• MCP over HTTP: a running SendMock also serves MCP at http://127.0.0.1:8025/mcp.
• The MCP registry: from 1.0, clients that browse the official MCP registry find SendMock as io.github.shibbirweb/sendmock, run through the npm package or the Docker image.

Reference:

• MCP reference: every tool, its inputs, an example call and what it returns.
• CLI reference: every command, the same way.

Three ways inCopy Anchor Link

Tool sets keep the agent's tool list small: the default is core, email, sms, verify, extract and webhooks; add payments, templates, variables, http, logs, admin or scenario with sendmock mcp --toolsets ..., or --toolsets all.

SafetyCopy Anchor Link

• Asks before destructive actions: reset, delete-all and deleting definitions are marked destructive, so your agent asks you first. sendmock mcp --read-only exposes only read tools.
• Never reads secrets back: agents can set a secret (an auth token, a gateway's secret key) but every answer shows it masked.
• Shows what the agent made: endpoints, rules and definitions created through MCP or the CLI carry a small badge in the UI, so you can review them.
• Stays local: webhooks go to local URLs unless you ask for others, and release to real SMTP still needs your allowlist.

NextCopy Anchor Link

• Build it with AI: prompts that work, for everything you can create.
• Create a gateway with AI: a complete worked example.
• Built-in prompts