MCP

Connect Claude, ChatGPT, Claude Code, Codex, Cursor, VS Code, or another AI client to Beai.

Beai serves a Model Context Protocol endpoint at /api/mcp, so an AI client can read your visibility data and manage prompts without you copying numbers between windows.

On cloud that is https://app.elmohq.com/api/mcp. A self-hosted instance serves the same endpoint on its own address, so swap app.elmohq.com for yours in everything below.

Every deployment also documents its own connection at Settings → MCP in the organization's settings sidebar: the endpoint on that instance's own address, snippets for the common clients, and the tools it registers.

There are two ways to connect. Signing in from the client is the quickest, and the connection then acts as you, in the workspaces you belong to. An API key is for a client that can't open a browser — a container, a scheduled job — and is also how you grant a client less than your own access.

Quick setup

add-mcp finds the AI clients installed on your machine and adds Beai to each of them:

npx add-mcp https://app.elmohq.com/api/mcp --name elmo

Add -g to install it for every project rather than just the current one, and -y to skip the prompt. Each client then asks you to sign in the first time it uses Beai. To set up a single client by hand, or one add-mcp doesn't cover, follow its section below.

Give the client the URL and it opens a browser for you to sign in. There is no application to register first, on cloud or on your own instance.

Claude Code

claude mcp add --transport http beai https://app.elmohq.com/api/mcp

Then run /mcp and choose Authenticate.

Claude.ai and Claude Desktop

Custom connectors are available on every Claude plan; the Free plan is limited to one.

Open Customize → Connectors, choose + Add, then Add custom connector.

Name it Beai, set the URL to https://app.elmohq.com/api/mcp, and choose Continue.

Review the detected authentication settings, choose Continue, and choose Add.

On Team and Enterprise plans, an owner adds the connector first under Organization settings → Connectors, choosing Add, then Custom → Web. Members then find it under Customize → Connectors and choose Connect.

ChatGPT

Custom MCP servers are added from ChatGPT on the web, and workspace admins can restrict them.

Open Plugins, choose +, then Add custom MCP server.

Name it Beai, set the server URL to https://app.elmohq.com/api/mcp, and choose OAuth for authentication.

Confirm the warning with I understand and want to continue, then choose Create as a plugin.

Install the plugin and sign in to Beai.

Codex

codex mcp add elmo --url https://app.elmohq.com/api/mcp
codex mcp login beai

Cursor

Add to Cursor

Or put it in ~/.cursor/mcp.json, or .cursor/mcp.json for a single project:

{
  "mcpServers": {
    "elmo": { "url": "https://app.elmohq.com/api/mcp" }
  }
}

Cursor shows the server as needing login; sign in from there. Older Cursor releases send a callback Beai can't accept, so if signing in fails, use an API key instead.

VS Code

Add to VS Code

Or from a terminal:

code --add-mcp '{"name":"elmo","type":"http","url":"https://app.elmohq.com/api/mcp"}'

Or run MCP: Add Server from the Command Palette, choose HTTP, and enter the URL. Or put it under servers in mcp.json:

{
  "servers": {
    "elmo": { "type": "http", "url": "https://app.elmohq.com/api/mcp" }
  }
}

VS Code asks you to sign in the first time it starts the server.

Devin Desktop

Devin Desktop, formerly Windsurf, shares its MCP servers with Devin CLI:

devin mcp add -s user elmo https://app.elmohq.com/api/mcp
devin mcp login beai

Or put it in ~/.config/devin/mcp_config.json:

{
  "mcpServers": {
    "elmo": { "url": "https://app.elmohq.com/api/mcp" }
  }
}

Antigravity

Antigravity's IDE and CLI, which replaced Gemini CLI, read ~/.gemini/config/mcp_config.json:

{
  "mcpServers": {
    "elmo": { "serverUrl": "https://app.elmohq.com/api/mcp" }
  }
}

To sign in, open Agent settings → Customizations and choose Authenticate next to Beai. After signing in, Antigravity shows a code in the browser; paste it back into the settings panel and choose Submit.

OpenCode

In opencode.json, or ~/.config/opencode/opencode.json for every project:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "elmo": {
      "type": "remote",
      "url": "https://app.elmohq.com/api/mcp",
      "enabled": true
    }
  }
}

Then opencode mcp auth elmo.

Goose

Add to Goose

Or run goose configure, choose Add Extension → Remote Extension (Streamable HTTP), and enter the URL.

Other clients

Any client that supports remote MCP servers over Streamable HTTP with OAuth can connect: give it https://app.elmohq.com/api/mcp and sign in when it asks. A client that only runs local servers can reach Beai through mcp-remote:

{
  "mcpServers": {
    "elmo": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://app.elmohq.com/api/mcp"]
    }
  }
}

Or use an API key

Issue one from Settings → API Keys in the organization's settings sidebar — keys belong to the organization, so they are not found under a brand. Issue it read-only and optionally narrow it to selected brands.

Send it as a bearer token. ChatGPT only connects by signing in; Claude.ai and Claude Desktop take the key when you add the connector: choose No sign in under Authentication and add an Authorization: Bearer elmo_... request header. add-mcp takes it as a header for every client it configures:

npx add-mcp https://app.elmohq.com/api/mcp --name elmo \
  --header "Authorization: Bearer elmo_..."

Claude Code:

claude mcp add --transport http beai https://app.elmohq.com/api/mcp \
  --header "Authorization: Bearer elmo_..."

Codex, which reads the key from the environment rather than the config file:

codex mcp add elmo --url https://app.elmohq.com/api/mcp \
  --bearer-token-env-var ELMO_API_KEY

Cursor, in ~/.cursor/mcp.json:

{
  "mcpServers": {
    "elmo": {
      "url": "https://app.elmohq.com/api/mcp",
      "headers": { "Authorization": "Bearer elmo_..." }
    }
  }
}

VS Code, in mcp.json:

{
  "servers": {
    "elmo": {
      "type": "http",
      "url": "https://app.elmohq.com/api/mcp",
      "headers": { "Authorization": "Bearer elmo_..." }
    }
  }
}

Devin Desktop and Devin CLI, in ~/.config/devin/mcp_config.json:

{
  "mcpServers": {
    "elmo": {
      "url": "https://app.elmohq.com/api/mcp",
      "headers": { "Authorization": "Bearer elmo_..." }
    }
  }
}

Antigravity, in ~/.gemini/config/mcp_config.json:

{
  "mcpServers": {
    "elmo": {
      "serverUrl": "https://app.elmohq.com/api/mcp",
      "headers": { "Authorization": "Bearer elmo_..." }
    }
  }
}

OpenCode, in opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "elmo": {
      "type": "remote",
      "url": "https://app.elmohq.com/api/mcp",
      "enabled": true,
      "headers": { "Authorization": "Bearer elmo_..." }
    }
  }
}

Self-hosting

Nothing has to be turned on — the endpoint is there as soon as the instance is, and signing in from a client works against your instance the same way it does against cloud.

The one thing that has to be right is APP_URL: it must be the address clients actually reach you on, because that is where the sign-in flow is advertised from and what the tokens it issues are bound to. It has to be an HTTPS URL, or localhost — the app refuses to start otherwise, since a token bound to anything else is one no MCP client will accept.

Keys come from the same organization settings page as on cloud, and instance-wide admin keys can also be configured server-side through the ADMIN_API_KEYS environment variable.

Try

What is the visibility and share of voice for each tagged prompt in Beai for <brand>?

What are some AEO opportunities for my <brand>?

Give me an example run for my <description> prompt.

For a full measure-diagnose-fix workflow on top of this data, install the skill.

Scopes

A connection is only offered the tools its scopes allow, so a client is never shown something it would then be refused:

Scope	MCP tool
(none)	`whoami`, `list_models`
`read`	`list_brands`, `get_brand`, `list_competitors`, `list_prompts`, `list_prompt_tags`, `get_analytics`, `get_prompt_performance`, `get_citations`, `get_query_fanout`, `get_opportunities`, `list_runs`, `get_run`, `get_billing`
`write`	`create_prompts`, `update_prompt`

A read-write key can create and edit prompts over MCP, but no tool deletes data and no tool changes billing — those are REST API concerns or reserved for an instance admin key entirely.

A connection you signed in from acts as you and reaches everything you can, so scopes are a property of a key rather than of the sign-in.

Was this page helpful?

MCP

Quick setupCopy Anchor Link

Sign in from the clientCopy Anchor Link

Claude CodeCopy Anchor Link

Claude.ai and Claude DesktopCopy Anchor Link

ChatGPTCopy Anchor Link

CodexCopy Anchor Link

CursorCopy Anchor Link

VS CodeCopy Anchor Link

Devin DesktopCopy Anchor Link

AntigravityCopy Anchor Link

OpenCodeCopy Anchor Link

GooseCopy Anchor Link

Other clientsCopy Anchor Link

Or use an API keyCopy Anchor Link

Self-hostingCopy Anchor Link

TryCopy Anchor Link

ScopesCopy Anchor Link