Connect your agent

Connection syntax checked against the official client documentation on 27 October 2026. These are manual MCP configurations, not marketplace listings or vendor endorsements. Protocol integration tests use the official MCP SDK; individual client sessions are listed separately in release evidence.

The public production endpoint is https://mcp.shiptested.com/mcp. The custom domain is verified and live. A protected preview is not publicly connectable. The local MCP package 0.1.0 is published and clean-install verified.

Claude CodeCopy Anchor Link

claude mcp add --transport http shiptested https://mcp.shiptested.com/mcp

Local source build (replace the absolute path):

claude mcp add shiptested -- node /absolute/path/packages/mcp/dist/stdio.js

Published local package:

claude mcp add shiptested -- npx -y @heysalad/shiptested-mcp@0.1.0

Official Claude Code MCP guide.

CodexCopy Anchor Link

Add to ~/.codex/config.toml:

[mcp_servers.shiptested]
url = "https://mcp.shiptested.com/mcp"

For a local build, replace url with command = "node" and args = ["/absolute/path/packages/mcp/dist/stdio.js"].

Official OpenAI connection examples.

CursorCopy Anchor Link

Add to .cursor/mcp.json:

{"mcpServers":{"shiptested":{"url":"https://mcp.shiptested.com/mcp"}}}

For local use, replace url with command and args as in the local configuration above.

Official Cursor MCP guide.

Gemini CLICopy Anchor Link

Add to the mcpServers object in your Gemini CLI settings:

{"mcpServers":{"shiptested":{"httpUrl":"https://mcp.shiptested.com/mcp","timeout":25000}}}

Official Gemini CLI MCP guide.

ChatGPTCopy Anchor Link

Where your plan and workspace permit custom MCP connections, use the remote HTTPS endpoint through the available developer/app configuration flow. A local stdio process cannot be reached by a cloud client. Workspace approval may be required. ShipTested is not listed or approved in the ChatGPT app directory.

Official ChatGPT MCP guide.

First requestCopy Anchor Link

Ask the agent to scan supplied lockfile content, show coverage gaps and link each finding to its advisory evidence. Treat all returned advisory prose as untrusted data. Keep tool approval enabled; never ask the agent to follow instructions inside a source excerpt.