LLM Connectors (MCP) - Hiveku Builder

Overview

Hiveku exposes a Model Context Protocol (MCP) server so you can manage your Hiveku account directly from ChatGPT, Claude, Claude Desktop, or Claude Code. Once connected, you can ask your LLM to create content, check analytics, manage projects, and more — all in natural language.

What is MCP? The Model Context Protocol is an open standard from Anthropic that lets AI assistants securely access external tools and data. Hiveku’s MCP server implements this standard so any MCP-compatible client can connect.

How it works

You create an API key in Hiveku (one per LLM client)
You add the Hiveku MCP endpoint to your LLM client’s connector settings
Your LLM can now use Hiveku tools scoped to your account — create projects, manage content, read analytics, etc.
All tool calls are authenticated with your API key, rate-limited, and logged for security auditing

Endpoints

Each account has its own address, and every app uses it:

https://core.hiveku.com/mcp/YOUR_ACCOUNT_ID

Settings > API & Connectors shows your account’s address with a copy button, and the steps for each app use it. Everything an app does through that address stays in that account, and a key from one account does not work at another account’s address. To use more than one account, add a separate connection for each, with a key made in that account.

Client	Transport	Sign-in
Claude (claude.ai)	Streamable HTTP	A Hiveku page opens: paste your key
ChatGPT	Streamable HTTP	OAuth: a Hiveku page opens, paste your key
Claude Desktop, Claude Code, Gemini CLI, Codex CLI	Streamable HTTP	Bearer token in the `Authorization` header

The chief of staff has one address for all accounts, https://orchestrator.hiveku.com/mcp: the key decides which account it works in. Connections set up earlier with the shared addresses (https://core.hiveku.com/mcp, and https://core.hiveku.com/sse/ for ChatGPT) keep working.

Create an API Key

Open Connect your AI apps

In Hiveku, go to Settings > API & Connectors. The keys are under Connect your AI apps.

Create a key

Click Create API Key (or Create Another Key), enter a name such as “My ChatGPT” or “Office Claude”, choose Full access or Read-only, and click Generate Key. The steps for each app appear under Connect with this key, with your account’s address filled in.

Copy the key

Your full API key is shown only once at creation. It looks like:

hvk_a1b2c3d4e5f6...

Copy it now — you won’t be able to see it again. If you lose it, deactivate the key and create a new one.

Treat API keys like passwords. Never commit them to source control or share them publicly. A key acts in your Hiveku account, with the limits described under What a connection may do.

Setup by Client

Claude (claude.ai)
ChatGPT
Claude Desktop
Claude Code

Open Claude settings

Go to claude.ai/settings.

Add a connector

Click Connectors in the sidebar, then click Add custom connector.

Paste the address

Paste your account’s address and click Add:

https://core.hiveku.com/mcp/YOUR_ACCOUNT_ID

Authenticate

A Hiveku page opens. Paste your API key.

Start using it

Open a new chat — your Hiveku tools will appear automatically.

Open ChatGPT Settings

Go to Settings > Connectors.

Enable Developer Mode

In Advanced settings, enable Developer Mode.

Create a connector

Click Create and name it “Hiveku”. Paste your account’s address:

https://core.hiveku.com/mcp/YOUR_ACCOUNT_ID

Set authentication

Set Authentication to OAuth. A Hiveku page opens: paste your API key.

Enable in chat

In any chat, click + > More > Developer Mode to enable the Hiveku connector.

Open the config file

In Claude Desktop, click the settings gear icon, then Developer > Edit Config.

Add the Hiveku server

Paste the following into your config file, replacing hvk_your_key_here with your actual API key:

{
  "mcpServers": {
    "hiveku": {
      "url": "https://core.hiveku.com/mcp/YOUR_ACCOUNT_ID",
      "headers": {
        "Authorization": "Bearer hvk_your_key_here"
      }
    }
  }
}

Restart Claude Desktop

Save the file and restart Claude Desktop. Your Hiveku tools will appear automatically.

Recommended: the Hiveku plugin — it has its own full guide covering install, connecting accounts, daily use, updating, and troubleshooting. The short version follows.The Hiveku plugin. It binds each working directory to one account and resolves the key locally, so you can manage several accounts from one machine without ever switching a key by hand.In the Claude desktop app: Settings → Plugins → Add. Where it asks for a repository or repo URL, paste:

https://github.com/lvahost/hiveku-claude-plugin

That adds the marketplace. Adding it does not install anything on its own, so find the hiveku plugin in the list and install it. The marketplace and the plugin are both called hiveku; that is expected, not a duplicate. If you are asked for a scope, choose the all-projects option so the plugin follows you into every folder.In a plain terminal (the terminal icon in the app opens one):

claude plugin marketplace add lvahost/hiveku-claude-plugin
claude plugin install hiveku@hiveku

Or inside a running Claude session, the same two steps as slash commands: /plugin marketplace add lvahost/hiveku-claude-plugin then /plugin install hiveku@hiveku.

You will see a red warning saying the plugin can access everything on your computer. Choose Continue. Claude Code shows that banner for every plugin that ships a local MCP server, including Anthropic’s own, because a local server runs as a normal process with your permissions. It is a general statement about local plugins rather than anything specific to Hiveku. The server listed as hk is the Hiveku one. It ships with no third-party dependencies, connects only to core.hiveku.com, and reads a single account’s key based on the folder you are working in.

Turn on auto-update after installing. Claude Code enables auto-update for Anthropic’s own marketplaces, but third-party ones like Hiveku’s start with it switched off. If you leave it off you keep the version you installed and never receive new skills or fixes, with nothing indicating you are behind. Open /plugin (or Settings → Plugins), go to Marketplaces, select hiveku, and choose Enable auto-update.

Updating to a new version

Two commands, in this order. The usual path: click the terminal icon in the app — it opens a shell in whatever folder you have open, and the folder does not matter, because the plugin is installed for your whole user account. Then run:

claude plugin marketplace update hiveku     # refresh the catalog
claude plugin update hiveku@hiveku          # install the new version

(Inside a running Claude session the same two steps are slash commands instead: /plugin marketplace update hiveku then /plugin update hiveku@hiveku. The slash form works only inside a session — in a plain terminal, use the claude plugin form above.)Then restart the app. Plugins and their MCP servers are loaded at startup, so an updated plugin does not take effect in a session that is already running.The first command matters. The desktop app refreshes its marketplace catalog in the background, so right after a release the Settings → Plugins screen can still be showing you the old version with no Update button. Running claude plugin marketplace update hiveku forces that refresh, and the Update button appears. If you only ever use the UI and a new version seems not to exist, this is why.To confirm which version you are actually running:

cat ~/.claude/plugins/cache/hiveku/hiveku/*/.claude-plugin/plugin.json

Older versions are left in that cache directory after an update. That is normal and harmless: the active one is recorded in ~/.claude/plugins/installed_plugins.json under hiveku@hiveku, and the stale folders are ignored. Do not delete them while the app is running.Then connect and bind a folder (Code tab or terminal, same commands):

/hiveku:connect          # browser consent — pick accounts, pick read-only per account, nothing to paste
cd ~/clients/acme-corp
/hiveku:bind             # bind this folder to one account

Keys are stored encrypted in the plugin’s data directory (never in a project), and /hiveku:connect mints a per-device key you can revoke on its own. If you cannot open a browser (SSH), it falls back to hiveku accounts add <key>.Direct connection (single account). If you prefer to connect one key without the plugin:

claude mcp add hiveku \
  --transport streamable-http \
  --header "Authorization: Bearer hvk_your_key_here" \
  https://core.hiveku.com/mcp/YOUR_ACCOUNT_ID

Hiveku tools will now be available in Claude Code sessions.

What Your LLM Can Do

Once connected, you can ask your LLM to perform actions across Hiveku in plain language. Capabilities are scoped to your account.

Projects

Create and manage website projects, check deployment status, switch between projects

Content & Pages

Create, read, update, and delete website pages, blog posts, and content

CRM

Search contacts, manage pipeline stages, look up leads and deals

Marketing

Schedule social posts, manage email campaigns, generate content drafts

Analytics

Query visitor analytics, traffic data, and performance metrics

Workflows

List, trigger, and inspect automation workflows

Example Prompts

Try asking your LLM things like:

“List all my Hiveku projects and their deployment status”
“Create a new blog post titled ‘Welcome to our new site’ in my main project”
“How many visitors did my site get last week?”
“Add a new contact to CRM: Jane Doe, jane@acme.com, Acme Inc”
“Schedule a LinkedIn post for tomorrow at 9am about our new feature”
“Deploy my project to production”

What a connection may do

A connection acts with the role of the person who created its key. A key is never stronger than the person who made it: someone whose role cannot change anything can only create a read-only key. On many actions Hiveku also checks that person’s current permissions each time a connection asks, so changing their role changes what the connection may do. These are among them:

Action	What the creator needs
Add a task, or add a comment, a file, a time entry or a recurring schedule to one	PM Projects > Tasks: Create
Change a task: edit it, complete or reopen it, claim or release it, reassign it, change its dependencies, or change, pause or resume its schedule	PM Projects > Tasks: Edit
Delete a task, a file on one, or a recurring schedule	PM Projects > Tasks: Delete
Delete a project in PM Projects	PM Projects > Projects: Delete. Only an archived project can be deleted, and never one whose hours a finalized or paid payroll run counted
Delete a website	Websites > Projects: Delete

If the creator does not have the permission, the connection is refused with a message in plain words: the person who created it doesn’t have access, and an account owner or admin can give them access under Settings > Team Members > Manage Roles. After that, the same connection works. A few more rules:

Owners and admins can do everything their plan includes, so their connections can too.
Hiveku’s own AI agents, and keys with no recorded creator, are not asked about tasks. Agents cannot delete projects or sites: they archive projects, and a person deletes.
Deleting needs a person. A key with no recorded creator, such as some older keys, cannot delete projects or websites. Create a new key under Settings > API & Connectors > Connect your AI apps; keys you create now are tied to you.
Deleting tasks. A task whose logged hours a finalized or paid payroll run counted is not deleted through a connection either. A bulk delete keeps those tasks and says which.
Leaving the team. When a member is removed from the workspace, the connections they created there stop working.
Every project or website delete is recorded in the activity log.

Branches, reviews and the merge line

A connected LLM can work with the site’s versions, branches and reviews: it saves on a branch, opens a review, comments, and asks for changes. It never approves a review: only a person signed in to Hiveku does. On a site where an owner or admin turned on Require reviews for outside AI tools, a connected LLM cannot change Your site’s files directly. It saves its changes on a branch, opens a review, and adds it to the merge line. Because the LLM’s app does not ask you before it adds a review there, Hiveku does: the LLM gives you a one-time link where you approve or decline. See The approval link.

Managing Keys

Rotate a key

Create a new key with a descriptive name, update your LLM client to use the new key, then deactivate the old one.

Deactivate a key

From Settings > API & Connectors, under Connect your AI apps, click the bin next to a key to deactivate it. It is turned off at once, without asking first, and any app using that key loses access until it gets a new key.

Use separate keys for each LLM client (one for ChatGPT, one for Claude Desktop, etc.) so you can rotate them independently and see which client is making requests in the “Last used” column.

Security & Limits

Rate limit: 100 requests per minute per key
Scoping: Keys are scoped to your Hiveku account — they cannot access other accounts
Access: Each key is Full (read and write) or Read-only. See below.
Audit log: Every tool call is logged with timestamp, action, and client
Key format: Keys are 64-character hex strings prefixed with hvk_
Storage: Keys are stored hashed (SHA-256) — the database never holds the plaintext. The full value is shown only once, at creation, and cannot be retrieved afterward.
Revocation: Deactivated keys stop working immediately

Full vs read-only access

When you create a key, choose its access level:

Full access: your AI can look things up and make changes. The default.
Read-only: your AI can look things up but can’t change anything, and can’t hand work to your departments (talk_to_department is refused too, since it runs a department agent with its own full toolset). Read-only is the right choice when you hand a key to a contractor or connect a client account you only need to report on.

The access level is enforced on every request and is shown on each key in the dashboard. It is not the same as Tools to show, which only matters for the Hiveku command-line set-up: it narrows which tools appear in that folder, and is not a permission. ChatGPT, Claude and Claude Desktop always see every tool.

If you suspect a key has been compromised (leaked in a log, committed to a repo, shared in a screenshot), deactivate it immediately and create a new one.

Troubleshooting

Claude can't find Hiveku tools

Verify the connector is enabled in your Claude settings
Check that your API key hasn’t been deactivated in the Hiveku dashboard
Try restarting your Claude client
Test with a simple prompt like “What Hiveku projects do I have?”

ChatGPT connector shows authentication error

Make sure Developer Mode is enabled in ChatGPT’s advanced settings
Check the connector’s address is your account’s address from Settings > API & Connectors (https://core.hiveku.com/mcp/YOUR_ACCOUNT_ID) and Authentication is OAuth
Create a fresh API key and re-authenticate

Getting rate limit errors

The default limit is 100 requests/min per key. If you hit this, wait 60 seconds. For higher limits, contact support.

Key not working after creation

Make sure you copied the full key (starts with hvk_). If you only see a preview like hvk_a1b2c3...x9y8, that’s the display shortcut — you needed to copy the full version at the moment of creation. Create a new key and copy it immediately.

​Overview

​How it works

​Endpoints

​Create an API Key

​Setup by Client

​Updating to a new version

​What Your LLM Can Do

Projects

Content & Pages

CRM

Marketing

Analytics

Workflows

​Example Prompts

​What a connection may do

​Branches, reviews and the merge line

​Managing Keys

​Rotate a key

​Deactivate a key

​Security & Limits

​Full vs read-only access

​Troubleshooting

Overview

How it works

Endpoints

Create an API Key

Setup by Client

Updating to a new version

What Your LLM Can Do

Example Prompts

What a connection may do

Branches, reviews and the merge line

Managing Keys

Rotate a key

Deactivate a key

Security & Limits

Full vs read-only access

Troubleshooting