MCP Server - Outbox AI

Overview

The Outbox AI MCP Server is a Model Context Protocol server that exposes your Outbox workspace as a set of tools any MCP-compatible client (Claude, Claude Code, Codex, Cursor, etc.) can call directly. Once connected, the model can list your agents, run tools, send messages, manage workflows, query CRM opportunities, fetch transcripts, and more — without leaving the chat.

The MCP server is brand-neutral. Whitelabel deployments can rename the server, tools, and responses without leaking the underlying provider’s identity.

How It Works

The server is a thin authenticated proxy over the Outbox HTTP API. Every MCP tool call translates into a single REST request scoped to your active client, or to your agency for agency operations.

Curated tools

A small set of tools is exposed by default for accurate model selection: list_records / get_record / create_record / update_record / delete_record, send_sms_message, send_email_message, send_ai_call, add_contacts_to_workflow, search_conversations, use_client and jarvis_query.

Every operation via run_operation

Everything else is reachable through run_operation with an operation name from list_operations (for example agents.duplicate or email_campaigns.launch). describe_operation shows its arguments.

Confirmation for destructive actions

Deleting, bulk sending, launching a campaign, enrolling more than five contacts, buying or releasing numbers and changing a client’s billing do not run on the first call. The server answers with needs_confirmation, a description of the effect and a confirm_token. The model shows you what will happen; only after you agree does it repeat the call with the token. A token works for 15 minutes and only for the exact same arguments, client and session.

Connection

The hosted server is at:

https://api.getoutbox.ai/mcp

Claude app
Claude Code (CLI)
Codex
Other clients

Works in Claude on the web and the desktop app. Once added, it’s available on mobile too.

Add the connector

Go to Settings → Connectors → Add custom connector. Name it Outbox, paste https://api.getoutbox.ai/mcp as the server URL, leave Advanced settings empty, and click Add.

Click Connect and sign in with your Outbox email and password. You’re returned to Claude, connected.

Turn it on in a chat

In a chat, open the tools menu under the message box and make sure Outbox is switched on.

On a Team or Enterprise plan, an owner adds the connector under Organization settings → Connectors first, then each member clicks Connect.

Add the server:

claude mcp add --transport http outbox https://api.getoutbox.ai/mcp

Then start Claude Code, run /mcp, choose outbox and select Authenticate. Sign in in the browser window that opens, then return to the terminal.

--transport http is MCP’s Streamable HTTP transport, not unencrypted HTTP. The connection is HTTPS because the URL is.

To make it available in every project rather than just the current one, add --scope user before the name.

Works in the Codex CLI, the Codex app and the Codex IDE extension, which share one setup. Add the server and sign in:

codex mcp add outbox --url https://api.getoutbox.ai/mcp
codex mcp login outbox

Sign in in the browser window that opens within 5 minutes, then return to the terminal. In the app or IDE extension you can instead add it under MCP servers → Add server as Streamable HTTP and click Authenticate.See Connect to Codex for config.toml and troubleshooting.

Clients that can’t sign in through a browser can authenticate with an API key instead. Generate one under Settings → Integrations in the dashboard, then send these headers on every connection:

Header	Value
`Authorization`	`Bearer <your-api-key>`
`X-Company-ID`	Your company UUID
`X-Agency-ID`	(optional) Agency UUID for agency-scoped access

For example, in Cursor or VS Code:

{
  "mcp": {
    "servers": {
      "outbox": {
        "type": "http",
        "url": "https://api.getoutbox.ai/mcp",
        "headers": {
          "Authorization": "Bearer YOUR_API_KEY",
          "X-Company-ID": "YOUR_COMPANY_ID"
        }
      }
    }
  }
}

See Connect to Claude or Connect to Codex for what you can ask, agency accounts, security, and troubleshooting.

Available Tool Groups

Group	Operations
Agents	List, create, get, update, delete, duplicate, test chat, prompt history and revert, AI prompt rewrite, tools, knowledge files, voices, calendars
Contacts	List, create, update, delete, bulk delete, import (CSV text or rows), export
Conversations	List, fetch by contact, call and chat threads, pause or resume the AI on a thread
Messages	Send SMS, email, Facebook and Instagram messages; bulk SMS and email
Calls	Trigger an AI call; view, pause, delete and bulk-queue calls
Workflows	List, create, get, update, delete, duplicate, folders, enrollments, execution logs, advance or remove contacts at a step, inbound webhook URL
Email Campaigns	Campaigns, steps, senders, enrollments, launch, logs, suppressions, sending domains and sender verification
CRM	Pipelines, stages and opportunities
Tools	Catalog, instances, configuration, schedules, run now, run logs, agent MCP servers
Integrations	Connect an app (returns a sign-in link for OAuth apps), status, disconnect
Phone numbers	Client numbers, search and buy numbers, order status and cancel, compliance checks, verified caller IDs
Settings	Client settings, staff and invites, custom fields and values, billing and subscription, connected AI apps
Websites	List, view, deployments, publish, revert
Agency	Clients (create, update, delete, assign plan), client plans, agency admins and invites, billing, white-label

Agency operations act on the agency rather than a client and need an agency admin. If you belong to more than one agency, the model passes target_agency_id (from list_accessible_companies). For the full list, call the list_operations tool from any MCP client once connected.

Self-Hosting

If you run your own backend, the server ships in api/mcp/company_server.py:

python3 -m api.mcp.company_server

Environment Variables

Variable	Default	Description
`MCP_BACKEND_BASE_URL`	`http://127.0.0.1:8000`	Outbox API base URL
`MCP_BACKEND_AUTHORIZATION`	—	Fallback `Authorization` header for stdio/internal use
`MCP_COMPANY_ID`	—	Company UUID forwarded as `X-Company-ID`
`MCP_AGENCY_ID`	—	Agency UUID forwarded as `X-Agency-ID` for agency operations
`MCP_CONFIRM_SECRET`	`SECRET_KEY`	Signs confirmation tokens. Set it when running more than one server process
`MCP_TRANSPORT`	`streamable-http`	FastMCP transport
`MCP_HOST`	`127.0.0.1`	HTTP bind host
`MCP_PORT`	`9001`	HTTP bind port
`MCP_PATH`	`/mcp`	HTTP mount path
`MCP_SERVER_NAME`	`platform-backend`	Server name advertised to clients
`MCP_REQUEST_TIMEOUT_SECONDS`	`45`	Per-request timeout
`MCP_EXPOSE_ALL_OPERATIONS`	`0`	Set to `1` to register one tool per operation
`MCP_SESSION_FILE`	`~/.platform/sessions/<key>.json`	Override path to the active-client session file (one file per signed-in user by default)
`MCP_PUBLIC_BASE_URL`	—	Public origin used in the OAuth challenge. Defaults to the `X-Forwarded-Host` your proxy sends

Best Practices

Scope API keys to the smallest needed surface

Generate a dedicated API key for each MCP integration. Rotate keys when devices change hands.

Prefer the curated tool list

Leave MCP_EXPOSE_ALL_OPERATIONS off unless you specifically need every operation as its own tool — large tool lists hurt model selection accuracy.

Use run_operation for ad-hoc calls

For one-off operations the model can call run_operation with the operation name and JSON arguments instead of exposing every endpoint.

​Overview

​How It Works

Curated tools

Every operation via run_operation

​Confirmation for destructive actions

​Connection

​Available Tool Groups

​Self-Hosting

​Environment Variables

​Best Practices

Overview

How It Works

Confirmation for destructive actions

Connection

Available Tool Groups

Self-Hosting

Environment Variables

Best Practices