DocsReference

Published Cloud MCP (Advanced)

The advanced direct-cloud MCP path using `@coppermind/mcp`. Most users should use the local `coppermind mcp serve` path instead.

mcpcloudadvancedtools

Most users should use the local Coppermind MCP path:

{
  "mcpServers": {
    "coppermind": {
      "command": "coppermind",
      "args": ["mcp", "serve"]
    }
  }
}

That path is local-first and does not require an account or API key for day-one success.

When the published cloud MCP package makes sense

Use @coppermind/mcp only when you explicitly want a direct cloud-connected MCP server instead of the local runtime-backed one.

That means you are choosing:

  • account-based auth
  • API key auth
  • managed cloud routes
  • a more advanced setup path

Install

First set your API key:

coppermind api-key set <cm_live_xxx>

Then configure your MCP client:

{
  "mcpServers": {
    "coppermind": {
      "command": "npx",
      "args": ["-y", "@coppermind/mcp"],
      "env": {
        "COPPERMIND_API_KEY": "cm_live_xxx"
      }
    }
  }
}

Optional environment variables:

  • COPPERMIND_BASE_URL for a non-default gateway
  • COPPERMIND_USER_ID for logical scope inside the tenant

Tools exposed

ToolPurpose
memory_searchSemantic lookup
memory_ingestStore a memory episode
memory_getInspect the current record set
memory_statsGet counts and token estimates
memory_deleteRemove stored records

Minimal verification

After restarting the client, a minimal proof looks like:

  1. run memory_stats
  2. run memory_ingest
  3. run memory_search

If those succeed with the expected data, the MCP server is wired correctly.

Common failure patterns

  • 401 or invalid_token: your cloud credential is wrong
  • memory_backend_not_configured: hosted sync or BYODB is not ready
  • memory_backend_unavailable: the storage plane is unhealthy

If you do not want to think about those cloud concerns yet, use the local runtime-backed MCP path instead.

Where to go next