How to Connect Claude and ChatGPT to Your Email, Calendar, and Files

Step 1 — Connect your accounts in Lodestar

Open Lodestar and go to Settings > Connected Accounts. Connect Gmail or Outlook for email and calendar. Connect Google Drive, OneDrive, or SharePoint for files. If your team uses Slack or Teams, connect that too — Lodestar ingests message threads and surfaces them the same way it does email.

Lodestar begins ingesting immediately. The enrichment pipeline runs in the background: action items are extracted and linked to their source quotes, contact sentiment is calculated across threads, org charts are inferred from communication patterns. You do not configure any of this — it runs automatically and updates as new data arrives.

Encryption is per-user and your data is never used for model training. If you prefer to bring your own API key for the AI layer, that option is available under Settings > AI Configuration.

Step 2 — Enable the MCP server

Once your accounts are connected and the initial ingest has run, go to Settings > MCP Server and enable it. Lodestar generates a server URL and an authentication token. Copy both — you will need them in the next step.

The MCP server exposes the full context layer to any compatible AI client. It does not push data to the model continuously — it waits for the model to call a tool and responds with exactly what the tool requests. Your data does not leave Lodestar until the model asks for it during a conversation.

Step 3 — Add Lodestar to Claude or ChatGPT

For Claude Desktop: open the app's settings, navigate to Developer > MCP Servers, and add a new server entry with the URL and token from the previous step. Name it "Lodestar" so it is easy to identify. Restart Claude Desktop and the tools will appear in the available-tools panel.

For Claude Code: add the Lodestar server to your .claude/mcp.json configuration file using the same URL and token. The server becomes available inside Claude Code sessions immediately — useful when you want to check your day's priorities, pull a spec from your knowledge graph, or close out a task without leaving the editor.

For ChatGPT: the exact setup depends on which ChatGPT interface you are using. Lodestar's docs (docs.getlodestar.ai > MCP > ChatGPT) have the current step-by-step — the general pattern is adding a custom action or plugin endpoint pointing to the Lodestar MCP server URL with your token as the authentication header.

Step 4 — Test with a real question

Open Claude and ask: "What needs my attention today?" If the connection is working, Claude will call the dashboard_today and needs_attention tools, pull your real data, and return a prioritised summary of your day — emails waiting on you, overdue tasks, upcoming meetings — all from your actual inbox and calendar.

Try a follow-up: "Draft a reply to the most urgent thread." Claude reads the thread content through emails_content, drafts a contextual reply, and surfaces it via emails_reply_link. You review it and click send. The reply is based on the actual conversation history, not a generic template.

The difference from a standard Claude session is immediately apparent. The model already knows your situation. You ask what to do next — not explain why.

What you can do once it is connected

With Claude connected to Lodestar, a representative day looks like this: morning briefing from your real data, triage of overnight email with AI-classified categories, a one-page meeting brief generated seconds before you join a call, action items captured from meeting notes and assigned to the right project, follow-up drafts ready to review at end of day.

None of these require switching apps or rebuilding context. The Sense-Decide-Act-Close loop runs through the same conversation window where you already work with Claude.

• ·"What is the status of the Meridian project?" — pulls project, action items, and last email thread
• ·"Prep me for my 2pm with Yolanda" — reads calendar event, org chart, open items, and prior threads
• ·"Create a follow-up task for the invoice conversation" — writes an action item linked to the source email
• ·"Mark the Q2 roadmap milestone complete" — updates the milestone and logs a timeline event
• ·"Which clients need attention this week?" — reads sentiment trends and surfaces the flagged relationships

Security and control

Every data access through the MCP server is scoped to your Lodestar workspace. The model cannot read data from other users or workspaces. Outbound actions — replies, task creation, milestone updates — are all logged to your timeline so you have a full audit trail of what the AI did on your behalf.

Email drafts go through a review step by design: Lodestar creates the draft in your mail client, and you choose whether to send it. The AI does not send email without your explicit action. This is a deliberate product decision, not a technical limitation.

Frequently asked questions