AI Tool Connections (MCP)

AI tool connections use the Model Context Protocol (MCP) to let supported assistants work with a single website after you sign in or approve access. Hiveality runs an MCP server at /api/mcp, so any MCP-compatible client can connect. Each connection is scoped to one website and gives the assistant the same set of tools no matter how you authenticate.

Supported assistants

Any Model Context Protocol (MCP) client can connect. Popular MCP clients include Claude, ChatGPT, Google Gemini, Cursor, VS Code, Codex, Antigravity, and Manus.

There are two connection methods. Claude and ChatGPT sign you in (OAuth). Codex, Manus, and most other clients use a Website API Key as a Bearer token. Both reach the same tools, scoped to this one website.

What a connected assistant can help with

After approval, a connected assistant can help with supported website tasks such as:

• Planning and starting website, funnel, and flow builds.
• Creating and updating blog posts, emails, campaigns, quizzes, contacts, boards, and bookings.
• Building, rebuilding, editing, and result-checking quizzes when you provide a complete brief or ask to revise an existing quiz. When you do not ask for a rename, rebuilding can keep the existing quiz title.
• Creating or generating emails with personalization tokens like {first_name}; common double-brace token mistakes are cleaned up before saved content is used.
• Updating contact custom fields, additional profile details, and address details when you ask.
• Reading board stage names before creating or moving cards, and creating cards that follow a board's saved default setup.
• Resolving eligible teammates before assigning board cards and pre-checking named checklist items during card creation.
• Reviewing page outlines, making focused text or link edits, and publishing pages when you explicitly ask it to. Assistants can see linkable buttons, images, icons, shapes, logos, and text links, including empty CTA buttons that still need a destination. They can target individual paragraphs inside larger blocks, preserve inline links while updating nearby copy, set or clear widget links after review, fill or replace image widgets, and remove reviewed paragraphs or sections when you approve that change.
• Creating draft pages, checking the live header menu, importing approved clean HTML into editable page drafts, and reporting any missing images or review warnings before you publish.
• Reading or updating site-wide brand styles and identity assets after approval, including colors, typography, button styles, custom fonts, logo, favicon, and social share image.
• Adding flow actions that send contacts to connected practice-management tools when the connection is enabled.
• Reviewing SEO health, broken links, and safe metadata suggestions for pages and blog posts.
• Cleaning up blog categories and tags without deleting the posts themselves, creating podcast episodes with their categories and tags, and keeping team profile details current.
• Finding booking availability, creating bookings, and reviewing or updating booking confirmation, cancellation, reschedule, and reminder messages after you approve the change.
• Creating and maintaining store products, courses, communities, events, boards, funnels, and popups (see the task notes below).
• Generating images into Files for use in blog posts, emails, pages, and funnels.
• Searching Hiveality docs, reading full guides, and returning docs links or dashboard links when you ask how to use a feature.
• Reviewing page outlines, form wiring, and submission settings without opening every page or form manually.
• Reading business metrics, form submissions, funnel details, store orders, products, courses, videos, and community activity when those features are available on the connected website.

Your connection URL

Find it in Website Settings → Integrations → AI Tools (the card above opens it). It is your app or website domain with /api/mcp at the end:

https://your-site.hiveality.com/api/mcp

Copy that URL — you will paste it into your assistant.

Which method does your assistant use?

Method 1: sign-in (Claude and ChatGPT)

Claude

1. In Claude, open connector settings and add a custom connector (MCP server).
2. Paste your connection URL.
3. When prompted, sign in to Hiveality and approve access.
4. Choose the website the assistant should work with, then allow access to return to Claude.

ChatGPT (desktop and web)

ChatGPT connects custom tool servers through Developer mode, available on paid plans (Plus, Pro, Business, and Enterprise). ChatGPT supports sign-in (OAuth) only, so a Website API Key is not used here.

1. In ChatGPT, go to Settings → Apps & Connectors → Advanced settings and turn Developer mode on.
2. Go back to Settings → Apps & Connectors → Add new connector (Create).
3. Enter a name and description, then paste your connection URL.
4. Set Authentication to OAuth, confirm you trust the connector, and click Create.
5. Complete the Hiveality sign-in to approve access.

ChatGPT turns Memory off while Developer mode is on. That is expected behaviour.

Method 2: API key (Codex, Manus, and others)

Some tools cannot open the sign-in approval flow. For those, use a Website API Key as the Bearer token. Both methods reach the same tools, scoped to this one website.

1. Create a key from Reveal Website API Key (it looks like hv_ws_…). Treat it like a password, and store it only in tools you trust.

2. In your assistant, add a custom server with your connection URL (HTTPS).

3. For authentication, choose API key / Bearer token and paste the key:

   Authorization: Bearer hv_ws_your_key_here
   

No sign-in step is required.

Manus

Add Hiveality as a custom server (not a built-in connector), then set the authentication to your Website API Key.

OpenAI Codex (CLI)

Codex reads MCP servers from ~/.codex/config.toml (or a project-scoped .codex/config.toml). Add a streamable-HTTP server and point it at your key via an environment variable:

[mcp_servers.hiveality]
url = "https://your-site.hiveality.com/api/mcp"
bearer_token_env_var = "HIVEALITY_API_KEY"

Then set the variable and start Codex:

export HIVEALITY_API_KEY="hv_ws_your_key_here"
codex

The codex mcp add command is for local (stdio) servers only. For a remote URL, edit config.toml as shown above.

Sign-in vs. API key: what's the difference?

• Same tools, same data. Both methods expose the identical tool set, scoped to the one website.
• Token lifetime. Sign-in tokens expire and refresh automatically. A Website API Key stays valid until you revoke it, which makes it a better fit for always-on or automated assistants like Codex, Manus, or n8n.
• Revoking access. Disconnect the connector (sign-in) or revoke the key in Integrations at any time.

Website selection

The approval page shows websites you can access. Search by site or business name when the list is long. A connection (and an API key) is tied to one website. If you want an assistant to work with a different website, add a separate connector (and key) for that website, or remove the current connection and connect again.

Notes for specific tasks

Most tasks just work after approval, but a few have setup caveats worth knowing:

• Store products are created as a draft. Publish them in the Store editor (or ask the assistant to) when you are ready to sell. The price currency must already be enabled in Store Settings → currencies, and a product becomes purchasable once it has a price and your store has payments connected.
• Courses, modules, and lessons created by the assistant start hidden from the live course. Publishing the whole course also publishes its lessons, including ones still hidden after an import. To change one lesson only, ask the assistant to publish or hide that lesson instead.
• Pages can be created as drafts, filed into folders, renamed, moved, and published or unpublished. Review the returned image report and preview link before publishing an imported draft.
• Funnels still need a page as their first step. The assistant can add page or popup steps and remove steps while keeping the sequence valid.
• Popups can be drafted or built from approved HTML, rebuilt without duplicating sections, and have their position, size, animation, overlay, and trigger rules adjusted before you publish. For manual setup, see Popups.
• Booking notifications should be reviewed service by service. Ask the assistant to read the event type first so it can keep existing reminders, use booking-specific message tokens, and avoid changing delivery behavior unless you asked for it.
• Communities, events, and boards can be listed, created, updated, and moderated after review, including board card moves with stage-automation summaries before bulk changes.

If the assistant can't find a tool, or tools look out of date

Your assistant fetches Hiveality's tool list once, when it connects, then remembers it for the session. We add and improve tools over time, so occasionally an assistant works from an older list and a newer tool looks missing.

Two things to try, in order:

1. Ask again, plainly. Sometimes the assistant simply did not reach for the right tool. Naming what you want ("create a new blog post" rather than "can you blog") usually gets there.
2. Refresh the connection so the assistant re-reads the current tool list:
   • Claude / ChatGPT: open the connector's settings and reconnect, or remove and re-add the connector. A fresh connection always pulls the latest tools.
   • Codex / Manus / n8n and other API-key clients: start a new session, or remove and re-add the server. The list is re-fetched on the next fresh connection.

You do not need a new API key for this. The key keeps working; only the assistant's remembered list of tools needs refreshing.

Security

• Only approve assistant connections you recognize, and connect one website at a time so access stays focused.
• Your Website API Key grants full access to that website's data. Keep it secret, never commit it to code or share it publicly, and rotate or revoke it instantly from Website Settings → Integrations if it is ever exposed. Each key is scoped to a single website, so a leaked key can never affect your other sites.
• Review generated content, emails, bookings, and image assets before sending or publishing.
• Review generated page warnings before applying or publishing a draft, even when the assistant says the generation completed. For imported page drafts, review the returned image report and preview link first.
• Ask the assistant to show what it plans to change before publishing important pages or broad taxonomy cleanup.
• For setup or troubleshooting questions, ask the assistant to cite the docs it used so you can review the source guide directly.