Connect Your AI Agent to Heights Platform with MCP
The Heights Platform MCP connects an AI assistant you already use to your Heights account. Instead of copying information between apps, you can ask your agent to look up a student, review a course, or update a product description using the writing style and business context you have already shared with it.
MCP stands for Model Context Protocol. Think of it as a connection that lets your AI app work with another service. Your AI app is sometimes called the MCP client; Heights provides the connection to your courses, products, students, and other supported account features. You do not need to build or host a server yourself.
Want the simplest setup? Install Heights from the LatchLoop plugin marketplace. For other apps, follow the relevant instructions below.
- What you need before connecting
- Your Heights MCP connection details
- LatchLoop
- Claude Code
- ChatGPT plugin updates
- Hermes Agent
- OpenClaw
- Test your connection
- What your agent can do
- Example workflows
- Approvals and account changes
- Troubleshooting and disconnecting
What you need before connecting Heights MCP
- A Heights creator or administrator account. Use your own account, not a student's login. The connection uses the access associated with your API key; additional-author permissions and plan requirements still apply.
- Your Heights subdomain. If your Heights address is
academy.heightsplatform.com, your subdomain is justacademy. Even if students visit a custom domain, use your original Heights subdomain for this connection. - Your Heights API key. In your Creator Dashboard, open Account Settings → Custom Code and find API Key. Copy the key privately. It is not your Heights password or an API key from your AI provider.
- Your preferred AI app. Use a current version with MCP support. Your app's subscription or workspace administrator may control access to plugins and external connections.
Treat your API key like a password. Enter it only in trusted connection settings or private local configuration. Do not paste it into an AI conversation, support message, screenshot, shared document, or code repository. Connecting an agent can give it access to sensitive student and order information, so review that app's privacy and data-handling settings too.
Your Heights MCP connection details
For a manual connection, use the following values. Replace YOUR_SUBDOMAIN with the short subdomain from the previous section.
- Name: Heights Platform (or
heightswhere a short name is needed) - Server URL:
https://mcp.heightsplatform.com/accounts/YOUR_SUBDOMAIN/mcp - Connection type / transport: Streamable HTTP (called http in Claude Code)
- Authentication: Bearer token
- Token: your Heights API key
For example, an account at academy.heightsplatform.com uses:
https://mcp.heightsplatform.com/accounts/academy/mcp
If the app asks for a Bearer token, paste only the key. If it asks for a custom header instead, set the header name to Authorization and its value to Bearer YOUR_API_KEY, with one space after Bearer.
Do not put the API key in the URL. You do not need a separate API base URL, a local Heights server, or an OAuth client ID and secret for this API-key connection. Follow your app's section below rather than pasting one app's configuration into another.
Install Heights Platform MCP inside LatchLoop
LatchLoop is a coding and general knowledge-work agent. Its Heights marketplace plugin handles the MCP connection for you.
- Open the Plugins page in LatchLoop.
- Find Heights in the marketplace and select Install.
- Enter your Heights subdomain and API key in the plugin's connection settings, then save the connection.
- Choose which LatchLoop projects can access the plugin. Enable it for the project where you want to work with Heights; you do not have to make it available to every project.
- Review the plugin's approval settings. You can keep the defaults or override them to be more or less strict about which types of actions require your approval. When getting started, keep approval required for actions that change your Heights account.
- Open a task or conversation in an enabled project and try the read-only connection test.
If the agent cannot see Heights, check both the saved connection and the project's plugin access. Project access controls where LatchLoop can use the plugin; approval settings control whether an available action can run without asking you first.
Connect Heights MCP to Claude Code
These steps are for Claude Code, Anthropic's agent, using its terminal commands. If you have not installed it yet, start with the official Claude Code quickstart.
- Open a terminal in the folder where you use Claude Code. Run the setup command in the terminal, not in the Claude conversation.
- In the command below, replace
YOUR_SUBDOMAINandYOUR_API_KEYwith your Heights details. Keep the quotation marks around the header, then run the command.
claude mcp add --transport http --scope local heights https://mcp.heightsplatform.com/accounts/YOUR_SUBDOMAIN/mcp --header "Authorization: Bearer YOUR_API_KEY"
Keep this command private: it saves your API key in Claude Code's local configuration and may also remain in your terminal history. Do not use a shared or recorded terminal. For setups that must keep credentials out of command history, Claude Code also supports environment-variable references in MCP headers.
- Start Claude Code with
claude, or restart your existing session in that folder. - Enter
/mcpinside Claude Code and check the heights connection. From the terminal,claude mcp listalso shows connection status. - Enter
/permissionsto review the tool permission rules, then try the read-only connection test.
--scope local makes this connection private to you for the current project, rather than writing the key into a shared project configuration. If you intentionally want it available across all of your Claude Code projects, use --scope user instead.
Approvals in Claude Code: in Manual/default mode, tools that need permission prompt unless a saved rule already allows them. Auto mode can review actions without routine prompts, and remembered approvals or broader allow rules can also change this behavior. Do not assume that every Heights edit will always produce a new prompt. Use /permissions to review Allow, Ask, and Deny rules and remove broad automatic approvals if you want to review changes yourself. See Claude Code's permission settings.
Heights Platform plugin for ChatGPT
We are considering an official Heights Platform plugin for ChatGPT to make connecting your account easier. We will add the installation steps here when it is available. Check this section for updates on the official plugin.
Connect Heights MCP to Hermes Agent
Hermes Agent can connect directly to Heights over HTTP. This setup uses two local settings files. If someone manages Hermes for you, give them this section and enter the API key privately yourself. The paths below use Hermes' default profile; ~ means your home folder.
- Open or create
~/.hermes/.envin a text editor. Add the line below, replacingYOUR_API_KEYwith your Heights API key. Keep this file private.
HEIGHTS_API_KEY=YOUR_API_KEY
- Open
~/.hermes/config.yaml. Add the following connection, replacingYOUR_SUBDOMAIN. If the file already has anmcp_servers:section, add only theheights:entry beneath it. Keep your other servers and preserve the indentation.
mcp_servers:
heights:
url: "https://mcp.heightsplatform.com/accounts/YOUR_SUBDOMAIN/mcp"
headers:
Authorization: "Bearer ${HEIGHTS_API_KEY}"
trust: untrusted
- Save both files and restart Hermes. Start a conversation with
hermes chat. - Try the read-only connection test. To review which actions Hermes can use, run
hermes mcp configure heightsin your terminal. Restart the conversation or use/reload-mcpafter changing the selection.
Leave ${HEIGHTS_API_KEY} exactly as written in the configuration: Hermes reads the key from your private .env file. Do not add auth: oauth or switch this connection to SSE.
Approvals in Hermes: trust: untrusted is Hermes' name for requiring approval on write-capable MCP actions; it does not mean the Heights service is unsafe. Hermes otherwise defaults to full trust. Keep approvals enabled and avoid YOLO mode if you want confirmation before changes. Choosing a session or permanent approval can allow later calls to that action without another prompt.
For current configuration and profile options, see the Hermes MCP guide and MCP configuration reference.
Connect Heights MCP to OpenClaw
Current OpenClaw releases include an MCP connection page. If your installation does not show it, update OpenClaw or ask the person who manages it to help. Make these changes on the machine running your OpenClaw Gateway, which may be different from the computer where you open its chat.
- In the OpenClaw Control UI, open Settings → MCP → Add server.
- Name the server
heights, choose Streamable HTTP, and enterhttps://mcp.heightsplatform.com/accounts/YOUR_SUBDOMAIN/mcpwith your Heights subdomain. Save the server. - Store your key privately in the Gateway's
~/.openclaw/.envfile by addingHEIGHTS_API_KEY=YOUR_API_KEY, replacing the placeholder with your key. If you use a named profile or a custom installation folder, use that installation's environment file instead. - Back in Settings → MCP, use the scoped configuration editor to add the
Authorizationheader to the savedheightsentry. The relevant configuration should look like the example below. Merge it with existing settings; do not replace your other connections.
{
"mcp": {
"servers": {
"heights": {
"url": "https://mcp.heightsplatform.com/accounts/YOUR_SUBDOMAIN/mcp",
"transport": "streamable-http",
"headers": {
"Authorization": "Bearer ${HEIGHTS_API_KEY}"
}
}
}
}
}
- Save the configuration and restart the Gateway so it picks up the private environment value.
- On the Gateway machine, run
openclaw mcp doctor heights --probeto check the connection, then try the read-only connection test in a new conversation.
Keep ${HEIGHTS_API_KEY} as a reference in the configuration, not the key itself. For this connection, use the header rather than enabling OAuth or choosing Sign in.
Approvals in OpenClaw: behavior depends on the agent runtime and permission settings you use. Do not assume connecting an MCP server automatically enables write confirmations. Review session tool access and the approval policy for your runtime; disable actions you do not want the agent to use. OpenClaw's Codex sessions have separate MCP approval controls, and their full-permission posture does not prompt by default.
See OpenClaw's MCP connection and approval guide and private environment settings for the options in your version.
Test your Heights connection without changing anything
Start with a lookup rather than an edit. In an enabled project or conversation, ask:
Use the Heights connection to tell me which Heights account I am connected to, then find up to five of my courses or products. Show their names. Do not create, edit, publish, enroll, or change anything.
Approve the lookup if your client asks. Check that the account and results belong to the Heights program you intended to connect. An empty product list can be valid for a new account; the account lookup should still identify the account. Merely seeing a saved connection or a list of available actions is not the same as successfully reading account data.
Next, try a specific question such as “Find my Photography Basics course and summarize its description.” You do not need to memorize action names or supply record IDs; the agent can find the matching item. If there are several similar matches, ask it to confirm the right one before making changes.
What your agent can do in your Heights account
Your agent can combine information you have shared in its conversation, project, or connected workspace with information it reads from Heights. It does not automatically know everything from your other chats or apps: attach the relevant writing samples, plans, or notes when needed.
- Find and review your products. Search courses, challenges, digital products, offers, and projects; inspect their descriptions and supported settings.
- Keep product information up to date. Change supported titles, descriptions, pricing, visibility, publication, and access settings. Available settings depend on the product type.
- Help you support students. Look up a student, review course completion, assignment answers, and project submissions, and investigate relevant orders.
- Manage student access. Enroll students, grant or revoke supported course or offer access and roles, or reset progress when you explicitly request it. Revoking access and resetting progress are consequential changes, not just lookups.
- Organize community channels. Find channels and groups, create channels, and update channel names or descriptions. Channel privacy is determined by attached offers; it is not a simple privacy switch the agent can change.
Content editing and Heights AI workflows
The expanded Heights MCP integration also supports the workflows below. Availability depends on the version exposed by your connection and your Heights permissions. If your agent does not see one of these newer actions, refresh the connection's available actions and check with support rather than assuming that every feature in the API reference is already available in your client.
- Improve lessons and course organization. Read lesson content and existing transcripts, create or edit lessons and modules, and make focused changes without rewriting unrelated content. Reading a transcript does not generate a missing one.
- Refresh existing blog posts. Find draft, scheduled, or published posts; revise specific passages; update titles and descriptions; and publish, unpublish, or schedule existing posts with the appropriate permissions. This does not create new blog posts.
- Prepare email content. Create or edit eligible campaign drafts, read automations, and revise emails in an automation's draft version. Sending and scheduling campaigns, or activating an automation, remain separate steps in Heights.
- Update website copy. Change supported page settings and plain-text sections while preserving the page's design. Use the Heights page editor for layout changes, complex content, and anything the connection cannot safely edit.
- Prepare certificate designs. Create or update certificate templates, including supported text, colors, fonts, and presets. This does not issue certificates to students or upload signatures.
- Arrange a live session. Find eligible destinations, review schedules, and schedule a single meeting or livestream with a specific date and time zone. Scheduling may queue reminder emails; it does not start the call. Recurring sessions, rescheduling, and cancellation are not part of this connection.
- Ask Heights AI to build for you. On a paid Heights plan, your external agent can request an email automation draft or a PDF digital-product package using Heights AI's own building workflow. It can check progress and return links to review the result. An automation build does not activate or send emails; a new PDF product is unpublished unless publishing is explicitly requested and authorized.
The MCP is not a replacement for every Heights screen. It does not expose every setting, the full course/offer Product Generator, or deletion actions. For interactive planning and visual review inside your account, you can also use Heights AI 3.
Example workflows using your context and Heights data
A useful request explains the goal, points to the context the agent should use, and sets a boundary for changes. These examples assume the agent already has the referenced materials; if not, provide them first.
Refresh a course description in your writing style
Context: your agent already has your brand notes, audience description, and a few newsletters you wrote.
Use my brand notes and recent newsletters to review the description of my Photography Basics course in Heights. Read the current description first, then suggest a clearer version in my voice. Keep the course's actual promises, price, and access settings unchanged. Show me the before and after, and wait for my approval before saving.
- The agent finds the course and reads its current information rather than guessing from the title.
- It compares that information with your writing samples and proposes a revised description.
- After you approve the wording and any required client approval, it saves only the agreed change.
- It reads the course again to confirm what was saved and reports the result.
Investigate a student's access question before changing anything
Context: you have given the agent a student's support message and your policy for which purchases include a particular course.
Using the support message and our access policy, look up this student in Heights, review their relevant orders and course access, and explain why they may not see the course. Draft a reply for me. If an access change is needed, show me the student, course, and proposed change and wait for approval. Do not reset progress or send a message.
- The agent matches the student using the email in the message and checks the relevant account records.
- It compares the purchase and access information with your policy, explaining anything it cannot establish.
- It drafts a reply in the conversation. If you approve a supported access correction, it can make that correction and verify it.
Drafting a reply does not send it. An access change also does not refund an order or cancel a subscription.
Turn course material into a companion workbook
Context: your agent has your teaching approach, brand preferences, and notes about what students should practice. This workflow needs the expanded content and PDF-building actions and a paid Heights plan.
Review the first module of my Photography Basics course in Heights and use the teaching notes in this project to propose a companion workbook. Show me the exercises and structure first. After I approve, ask Heights AI to build a new PDF digital product, keep it unpublished, and give me the package-builder link when it is ready. Leave the existing lessons unchanged.
- The agent locates the course, reads the relevant lessons, and combines them with your teaching notes.
- It proposes exercises tied to the actual course material for you to review.
- Once approved, it requests the Heights AI build and checks its progress instead of reporting a queued job as finished.
- It returns the completed result's review links and publication status. You review the PDF in Heights before making it available to students.
Approvals and changes to your Heights account
A read looks up information. A write creates or changes something: saving a draft, changing a price, publishing a page, enrolling a student, or scheduling a call all count. Even an unpublished draft is a change to your account.
Your AI client controls approval prompts. Heights identifies actions as read-only or change-making to help the client apply its rules, but those labels do not guarantee a prompt. The client's mode, saved approvals, project settings, and administrator policies determine what it asks you to confirm. Once a permitted change-making action runs successfully, it can save immediately in Heights; there is not necessarily a second confirmation screen there.
In LatchLoop, choose which projects can access Heights and override the plugin's default approval settings to be more or less strict. In Claude Code, review /permissions and your session's permission mode. For Hermes and OpenClaw, use the approval guidance in their setup sections above.
- Start with lookups. Confirm the connected account before asking for edits.
- Review the target and effect. Check the student, product, proposed wording, price, publication state, and any time zone before approving an action.
- Be careful with “Always allow.” A remembered approval may cover future calls to the same action, including different students or products.
- Request a preview in the conversation. “Show me the proposed changes and wait” is useful guidance, but it is not a substitute for enforced client approval settings.
- Ask for a read-back. Have the agent verify saved results. Edits to already-public content can become visible immediately, even if you did not separately ask to publish it.
Some edits check whether content changed since the agent read it. If another edit happened in the meantime, ask the agent to read the latest version and review a new proposal. This protects against overwriting newer work; it is not a universal undo feature.
For supported blog content edits, the post's History shows Name · MCP and retains the five latest meaningful content changes. You can preview and restore content in the Heights editor. Restoring content does not revert publication settings, and other kinds of changes do not necessarily have the same recovery options.
Troubleshooting and disconnecting Heights MCP
- The connection cannot authenticate: check the original Heights subdomain, use an administrator's API key from that account, and remove accidental spaces or line breaks. In a bearer-token field use the key alone; in an Authorization header use
Bearerfollowed by one space and the key. Check that any private environment value is available to the process actually running your agent. - The connection works, but an action is denied: check Heights author permissions and plan requirements as well as your client's approval policy. Reconnecting cannot grant permissions your Heights user does not have. Heights AI automation and PDF builds require a paid account.
- The agent cannot see Heights or a newer action: check project access, enabled plugins, and action filters. Refresh the available actions or restart the conversation. Different integrations can expose different capabilities.
- A page or email cannot be edited: use the Heights editor for unsupported page content. Campaign edits require an eligible unscheduled or failed draft; automation email edits apply to the current draft, not the active version.
- A build is still running: ask the agent to check the existing build's status. If a save or build times out or its outcome is uncertain, inspect Heights before repeating it so you do not create duplicate work.
- Too many requests: pause and retry the lookup later. API limits are shared with your other integrations; avoid asking the agent to repeatedly scan your whole account.
To stop using the connection, disable or uninstall the Heights plugin in your client. In LatchLoop you can also remove access from a specific project. In Claude Code, run claude mcp remove heights from the project where you added it. In Hermes, disable the heights server; in OpenClaw, disable or remove it in Settings → MCP.
Removing a connection does not undo changes already made or invalidate copies of your Heights API key. If you believe the key was exposed, contact Heights support for help securing it and review your other integrations that use the same key. When contacting support, share the client name, the action you attempted, and a redacted error message—not the key or unnecessary student data.
Building your own integration? See the Heights Platform API documentation. For other ways to connect services, see API & Zapier Integration.