Connect your assistant

The MCP server speaks Streamable HTTP at POST https://mcp.aymaragents.com/mcp — configure your client as a remote/HTTP MCP server, not a local command/stdio one. The exact config block shape differs slightly per client; what’s constant across all three: URL = https://mcp.aymaragents.com/mcp, one header X-Api-Key with your full API key.

Verification

These blocks are verified against the server’s real protocol (handshake, tools list filtered by your key, confirm on writes) with the official MCP SDK client — verification with the real Claude Desktop, Claude Code and Cursor applications, with screenshots, is pending MCP-4. If one of these clients’ own UI has changed since this was written, its own MCP documentation is the final reference.

Pick your client

The exact config block shape differs slightly per client — each has its own page, with the full block and how to verify it:

• Claude Desktop
• Claude Code
• Cursor

A note on TLS

If the MCP server responds over https:// (any production deployment), your client is talking plain HTTPS to it — normal OS/browser trust applies, nothing to configure here. This is different from the outbound TLS that same server uses to talk to the public API — an infrastructure detail of ours, never something your MCP client needs to configure.

Before you connect

1. Confirm your plan is Pro or Enterprise (see Plans and limits).
2. Create or reuse an API key with the scopes you need (guide API keys) — the key is shown in full only once.
3. Pick your client above and paste that key into its config block, replacing <your full API key>.
4. Ask your assistant to call the aymar_status tool — if everything is right, you’ll see your tenant, your plan and your effective scopes. If something fails, continue with Limits, errors and troubleshooting.

Continue with the tools catalog to see what your assistant can do once connected.