Agent Skills: Remote MCP Integration

Use when configuring, testing, or implementing remote Clous MCP server integrations for Codex, Claude, Cursor, ChatGPT, or other MCP clients.

UncategorizedID: clous-ai/agents/remote-mcp-integration

Repository

clous-aiLicense: NOASSERTION

Install this agent skill to your local

pnpm dlx add-skill https://github.com/clous-ai/plugins/tree/HEAD/clous-development/skills/remote-mcp-integration

Skill Files

Browse the full folder contents for remote-mcp-integration.

Download Skill

Loading file tree…

clous-development/skills/remote-mcp-integration/SKILL.md

Skill Metadata

Name
remote-mcp-integration
Description
Use when configuring, testing, or implementing remote Clous MCP server integrations for Codex, Claude, Cursor, ChatGPT, or other MCP clients.

Remote MCP Integration

Clous publishes remote MCP servers rather than bundling local server binaries in this plugin.

Servers

  • clous-platform: https://mcp.clous.app/
  • clous-hr: https://mcp.clous.app/hr
  • clous-employee: https://mcp.clous.app/employee

Workflow

  1. Load .mcp.json from the plugin root or generate equivalent config with the SDK.
  2. Confirm the host supports remote HTTP MCP servers.
  3. Authenticate through the host's OAuth flow when required.
  4. Run discovery/list tools before invoking mutations.
  5. Inspect tool schemas, scopes, and resource exposure before wiring host prompts or agent actions.
  6. Treat UI resources, structured content, and conversational text as separate output channels.
  7. Preserve traceability: keep the MCP server, tool name, resource id, and host/client assumption visible in integration notes.

Safety

  • Do not pass raw employee data to unrelated tools.
  • Do not bulk export without destination, purpose, and user confirmation.
  • Confirm write tools have a response path back to the product or conversation.
  • Prefer least-privilege scopes and explain which people-data context each tool can read or mutate.
  • For HR or performance data, require citation or resource references in answers so the host does not present retrieved context as unsupported fact.