MCP Connector Skill (/cm-connect)

How to use the /cm-connect skill to pull data from any connected MCP into Coppermind's memory system.

What It Does#

/cm-connect bridges external data sources (Granola, Gmail, Slack, HubSpot, Google Calendar, or any MCP server connected to your Claude session) into Coppermind. It fetches data using whatever MCP tools are available, extracts discrete memories, and stores them in the active client mind.

This is the universal ingestion path. If an MCP exists for a data source, /cm-connect can pull from it.

Basic Usage#

Pull recent meetings from Granola#

/cm-connect Pull in my last 3 meetings with Acme from Granola

Import emails about a client#

/cm-connect Import the last 10 emails about Bluebell from Gmail

Refresh from a previously ingested source#

/cm-connect Refresh Acme's Granola meetings from this week

The "refresh" keyword tells the skill to re-process even if the data looks familiar. Reconsolidation handles merging with existing memories.

How It Works#

1. Identifies the target client mind. If you name a client ("add to Bluebell"), it switches to that client mind. If no client is named, it asks which client mind to use.

1. Fetches data. Uses available MCP tools in the session to retrieve what you described. If the required MCP is not connected, it tells you what to add.

1. Extracts discrete memories. Each piece of content is broken into standalone memories with explicit types:

• decision -- choices made
• preference -- stated preferences
• campaign_outcome -- results with metrics
• commitment -- action items with owners
• stakeholder -- people information
• fact -- general business context

1. Stores each memory. Calls store_memory for each extracted memory with the type set explicitly. Reconsolidation handles dedup automatically.

1. Reports results. Summarizes items retrieved, memories stored (new + updated), target client mind, and any skipped items.

Rate Limits#

Rules the Skill Follows#

• One memory per store_memory call. No combined facts.
• Concise. 1-3 sentences per memory, standalone, useful without context.
• Preserves specifics. Names, dates, numbers, exact commitments.
• Skips noise. Greetings, small talk, scheduling logistics.
• Never fabricates. Only stores what exists in the source data.

Examples by Data Source#

Granola (meeting transcripts)#

/cm-connect Pull all Granola meetings from this week for Acme

Extracts decisions, action items, stakeholder preferences, and campaign outcomes from meeting transcripts.

Gmail (email threads)#

/cm-connect Import emails from Sarah about the Q3 budget

Pulls relevant emails and extracts commitments, decisions, and context.

Google Calendar#

/cm-connect Check my calendar for Acme meetings this week and log them

Detects meetings and logs them for follow-up tracking.

HubSpot (CRM data)#

/cm-connect Pull the latest deal status for Bluebell from HubSpot

Extracts pipeline status, contact info, and recent activity.

Checking Available MCPs#

If you are unsure which MCP sources are connected in your current session, ask Claude:

What MCP servers are connected right now?

Claude will list the available MCPs. If the one you need is not connected, you will need to add it to your Claude Code or Claude Desktop configuration.

Related Guides#

• Auto-Ingest from Connected Sources -- automatic background ingestion
• Transcript Ingestion -- processing meeting transcripts
• Memory Storage and Search -- how stored memories work