Skip to content

Cursor

Ask about this page ChatGPT Gemini Claude

This connects Cursor to the Halyard MCP server so the agent can search knowledge, ask teammates, and capture answers. The plugin is the recommended path — it ships the MCP server plus the skills and an always-on rule that teach the agent when to search, ask, and log. Manual MCP config gives you the tools only.

Section titled “Option 1: Install the Halyard plugin (recommended)”

Install Halyard from the Cursor Marketplace — search for “Halyard”.

The plugin installs:

  • The org-kb MCP server (mcp.usehalyard.ai) — search, ask-expert, and knowledge-capture tools
  • Six skillsgather-context, ask-for-help, log-work, review-work, triage-knowledge, mark-terms
  • A knowledge-first rule that orients every session toward the knowledge base: search before substantive work, capture outcomes after

You can also install straight from the repository before (or without) the marketplace listing:

Terminal window
git clone https://github.com/halyard-labs/claude-plugin-marketplace
ln -s "$(pwd)/claude-plugin-marketplace/plugins/halyard" ~/.cursor/plugins/local/halyard

Add the Halyard server to your project’s .cursor/mcp.json:

.cursor/mcp.json
{
"mcpServers": {
"halyard": {
"url": "https://mcp.usehalyard.ai"
}
}
}

Reload Cursor after installing the plugin or editing the config. On its first call to a Halyard tool, Cursor opens a browser to authorize against your Halyard account. Approve it once; the bearer token is then attached to every request.

Ask the agent to identify you and run a search:

Use Halyard to tell me who I am, then search the knowledge base for our most recent decision.

Expect a whoami call returning your user and organization, followed by a search_knowledge call. If the tools do not appear, see Troubleshooting.