Mindsmith MCP: Connect Claude, Cursor, Copilot & ChatGPT

Mindsmith works with the AI tools you already use. Connect it once and you can ask Claude, Cursor, GitHub Copilot, ChatGPT, or any other MCP-compatible assistant to build lessons, pull engagement numbers, or generate tracked links — without leaving the chat.

This is done over the Model Context Protocol (MCP), an open standard that lets AI tools talk to apps like Mindsmith securely.

All you need

The Mindsmith MCP server URL. Use the one that matches your workspace region:

Region	URL
US (most accounts)	`https://app.mindsmith.ai/api/mcp`
EU	`https://eu.mindsmith.ai/api/mcp`

Not sure which region you're on? It's whichever you log in to in your browser today. Copy your URL — you'll paste it once below, sign in to Mindsmith in a popup, and you're done.

Add Mindsmith to your AI tool

Pick your tool below. The setup takes about 60 seconds.

Claude (web — claude.ai)

In the left sidebar, click Customize → Connectors (or go straight to claude.ai/settings/connectors).
Click the + next to Connectors and choose Add custom connector.
Paste your Mindsmith URL: https://app.mindsmith.ai/api/mcp
Click Add. A Mindsmith window will open — sign in, pick your workspace, and choose whether to allow editing. Click Authorize Access.

You're done. Try a prompt like "In Mindsmith, list my lessons in the Onboarding project."

Custom connectors are available on Pro, Max, Team, and Enterprise plans (free users get one).

Claude Desktop (Mac / Windows)

Same flow as the web app — in the sidebar, click Customize → Connectors → Add custom connector, paste https://app.mindsmith.ai/api/mcp, and authorize.

If you're on an older build that doesn't have the Connectors menu, open your config file:

Mac: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

Add this and restart Claude:

{   "mcpServers": {     "mindsmith": {       "url": "https://app.mindsmith.ai/api/mcp"     }   } }

Cursor

Open Cursor Settings → Tools & MCP (older builds: Features → MCP).
Click + Add Custom MCP Server.
Name it mindsmith, choose Streamable HTTP as the transport, and paste the URL.
Save. Cursor will open a Mindsmith window for you to sign in and authorize. A green dot next to the server name means it's connected.

If Cursor asks for JSON instead of fields:

{   "mcpServers": {     "mindsmith": {       "url": "https://app.mindsmith.ai/api/mcp"     }   } }

Microsoft Copilot (via Copilot Studio)

The consumer Copilot at copilot.microsoft.com currently only accepts a fixed set of built-in connectors (OneDrive, Outlook, Gmail, Google Drive, Google Calendar, Google Contacts) — you can't paste an arbitrary MCP URL into its Settings → Connectors page yet.

To use Mindsmith from Microsoft Copilot today, register it in Copilot Studio. The agent you build there becomes available in Microsoft 365 Copilot.

Open Copilot Studio and pick (or create) the agent you want Mindsmith available in.
Make sure generative orchestration is turned on for the agent — it's required for MCP.
Go to Tools → Add a tool → New tool → Model Context Protocol.
Fill in:
- Server name: Mindsmith
- Server description: Build, manage, and analyze Mindsmith lessons.
- Server URL: https://app.mindsmith.ai/api/mcp
Click Create. Copilot Studio generates a custom connector under the hood that streams to Mindsmith over HTTP. Authorize the connection when prompted.
Publish your agent so it shows up in Microsoft 365 Copilot.

You'll need a Copilot Studio license and Power Platform environment for this path.

ChatGPT

Custom MCP connectors in ChatGPT are in beta and require Developer Mode. They're available on Plus, Pro, Business, Enterprise, and Edu (web).

Turn on Developer Mode.
- Individual users: Settings → Apps → Advanced settings → Developer mode.
- Workspaces: an admin enables it under Workspace Settings → Permissions & Roles → Connected Data → Developer mode.
Go to Settings → Connectors → Create.
Give it a name (e.g. Mindsmith) and a short description, paste https://app.mindsmith.ai/api/mcp as the MCP Server URL, set Authentication to OAuth, check I trust this application, and click Create.
Sign in to Mindsmith when the popup opens.

To use it in a chat, click the + in the message bar → Developer Mode → Add sources and turn on Mindsmith.

Claude Code (CLI)

claude mcp add --transport http mindsmith https://app.mindsmith.ai/api/mcp

The first time you ask Claude Code to do something with Mindsmith, it will open the sign-in page for you. Add -s user to install it globally for your user, or -s project to commit a .mcp.json for your team.

Windsurf, Zed, and other MCP clients

Any tool that supports remote MCP servers can connect. Wherever the tool asks for an MCP server URL, paste https://app.mindsmith.ai/api/mcp (or the EU URL). Sign-in happens in your browser.

Windsurf: Open Windsurf Settings → Cascade → MCP Servers → Add Custom Server, or edit mcp_config.json and add a serverUrl entry.
Zed: Open the Agent Panel (⌘⇧A), click the menu → View Server Extensions, or add a context_servers entry to settings.json.

Choose what the AI can do

When you sign in, Mindsmith will show a small consent screen. You'll:

Pick your workspace. If you belong to more than one Mindsmith workspace, choose the one you want this AI tool to use.
Decide on Write access.
- Read access is always on. The AI can browse your lessons, search them, and read your analytics.
- Write access is a toggle. Turn it on if you want the AI to create lessons, edit them, publish/unpublish, and generate tracked links. Leave it off if you only want look-but-don't-touch.

You can change this anytime — just connect again and pick differently.

Try it

Some prompts to try once you're connected:

"List my Mindsmith projects and tell me how many lessons are in each."
"Find the Mindsmith lesson with the highest completion rate this month."
"Create a Mindsmith lesson called 'Quarterly Security Refresher' covering password hygiene, phishing red flags, and how to report incidents. Put it in the Compliance project."
"For my onboarding lesson, show me where learners are dropping off."
"Generate tracked links for these 12 people and put them in a table I can paste into an email."

What the AI can do for you

Once connected, here's what your AI assistant can do in Mindsmith:

Look things up

Browse projects and lessons
Search across all your content
Open a specific lesson and see its full structure
Pull top-level results (views, completions, scores)
Drill into per-block engagement, including for individual learners

Make and share content (with Write access on)

Create a new lesson by describing what you want — the Mindsmith AI agent builds it
Edit an existing lesson the same way
Publish or unpublish a lesson
Generate individually tracked links for a list of recipients

Common questions

Is my data safe? Yes. Each connection is scoped to one workspace, uses standard OAuth (the same flow you'd use to sign into another app with Google), and you control read vs. write access. You can revoke access anytime from the AI client.

Do I need an API key? No. There's nothing to copy or paste. Authentication happens in your browser.

It says I'm not authorized after a while. Tokens refresh automatically, but if a client loses its credentials, just disconnect and reconnect — no harm done.

The AI says it can't edit lessons. Write access is a separate toggle. Reconnect and turn on Write access at the consent step.

My workspace isn't showing up. The consent screen lists every workspace your Mindsmith account belongs to. If one is missing, check that your invite is accepted and your membership is active.

Why does the connector reach Mindsmith from the cloud, not my computer? For claude.ai, Claude Desktop, and the Claude mobile apps, Anthropic connects to remote MCP servers from their cloud infrastructure. Mindsmith is a public HTTPS service, so this is fine — but it's why the URL must be reachable on the public internet.

For developers

If you're building a custom integration or want the protocol-level details, here's how the Mindsmith MCP server is wired up.

Endpoints

MCP endpoint (Streamable HTTP transport):
- US: https://app.mindsmith.ai/api/mcp
- EU: https://eu.mindsmith.ai/api/mcp
OAuth discovery (RFC 8414 / 9728):
- /.well-known/oauth-protected-resource
- /.well-known/oauth-authorization-server

Auth flow

The server implements the MCP authorization spec end-to-end:

OAuth 2.0 with PKCE. S256 only — plain is rejected.
Dynamic Client Registration (RFC 7591). Clients register at /api/oauth/register on first use. No developer console signup is required.
Public clients. token_endpoint_auth_methods_supported: ["none"].
Refresh tokens. Issued alongside access tokens.
Revocation. /api/oauth/revoke follows RFC 7009.

Scopes

Scope	Grants
`mcp:read`	Read tools (always granted)
`mcp:write`	Write tools (opt-in via consent toggle)

Every mcp:write action that publishes or unpublishes a lesson is recorded in the workspace audit log.

Tool reference

Ten tools are registered. Read tools are always available; write tools require mcp:write.

Read

Tool	Description
`list_projects`	List all projects in the organization with lesson counts.
`list_lessons`	List lessons in the organization. Filter by project, content type, or find unassigned lessons.
`search_lessons`	Search lessons by keyword across titles, descriptions, and full text content.
`view_lesson`	Get a lesson's status, URLs, and optionally full content. Use `includeContent=false` when polling for generation progress.
`get_lesson_results`	Top-level lesson results: unique viewers, starts, completions, completion rate, average score, and daily view trends.
`get_lesson_engagement`	Per-block engagement plus a list of learners with scores. Pass a `learnerId` to drill into a single learner's per-attempt detail.

Write

Tool	Description
`create_lesson`	Create a new lesson with the Mindsmith AI agent. Returns a lesson ID — poll `view_lesson` until `status: "ready"`. Optionally duplicate from `sourceLessonId`.
`edit_lesson`	Edit an existing lesson with the Mindsmith AI agent. Same async pattern as `create_lesson`.
`update_lesson_sharing`	Publish or unpublish a lesson.
`get_tracked_links`	Generate individually tracked links for a lesson, pre-filled with learner names and emails. Works for unpublished lessons.

Async generation

create_lesson and edit_lesson kick off a Mindsmith agent run and return immediately with a lesson ID. Poll view_lesson (with includeContent=false for cheap polls) until status: "ready". Most lessons finish in under a minute.

Limits

Individual MCP requests are capped at 30 seconds. Long-running work is async, so this should not be a factor in normal use.
Each connection is bound to one workspace. To use multiple workspaces, register the server multiple times in your client (e.g. mindsmith-acme, mindsmith-globex).

Getting Started: Generating a Lesson in Mindsmith