MCP.so
Sign In

OneVerifiedFeatured

@One

About One

Connect your agent to the apps your team already uses: Gmail, Slack, Stripe, Shopify, HubSpot, Notion, Linear, Salesforce, QuickBooks and more. One hosted MCP server with OAuth sign-in, nothing to install and no API keys to manage. Actions are discovered on demand along with their real API documentation, so your agent works from each platform's actual schema instead of a guess.

Connection details

https://mcp.withone.ai/mcp

Setup

claude mcp add one --transport http https://mcp.withone.ai/mcp

Tools

4

Use this to run an API action on a platform the user has connected on One, after `get_one_action_knowledge` documented its inputs. Takes the `actionId` from `search_one_platform_actions` (or the connection's `access` list) and the `connectionKey` from `list_one_integrations`. This can create, modify, delete, or send data on the external platform — confirm irreversible operations with the user before calling. Do not use it to probe connectivity or retry a failed write blindly. Requires the `user:connections:write` OAuth scope; a knowledge-only or access-rule grant can still refuse the call.

Use this immediately before executing an action: it fetches the action's knowledge document — required parameters, validation rules, and platform-specific caveats — enriched with guidance on how to call it. Never call `execute_one_action` without reading this first. Refused when the action is outside the caller's grant. Requires the `user:connections:read` OAuth scope.

Use this when you need to know which platforms the user has connected on One — and always before searching or executing. Lists the active connections with their `key` (the `connectionKey` for execution) and the `access` each grant confers: `full`, a set of allowed HTTP `methods`, or a specific list of `actions` (each with `actionId`, `title`, `method`). When a connection is action-scoped, its `actions` are exactly what may run, so you need not search. Do not guess connection keys — only use values returned here. Requires the `user:connections:read` OAuth scope.

Use this when you need to find an API action on a connected platform (e.g. "send an email" on gmail). Searches the platform's action catalog, scoped to the caller's grant: results are limited to actions the grant permits on that platform's connections. Returns up to 5 matching actions with their `actionId`, title, HTTP method, and path. Do not use this for action-scoped connections — their `access` list from `list_one_integrations` already names the permitted actions. Requires the `user:connections:read` OAuth scope.

Overview

What is One

One is a hosted MCP server that connects your agent to the apps your team already uses, through a single endpoint. Instead of installing one MCP server per app, you point your client at https://mcp.withone.ai/mcp once and sign in with OAuth.

How it works

Your agent lists the accounts you have connected, searches a platform for the action it needs, reads that action's real parameters and authentication before calling it, then executes it against the live account.

Because actions are discovered on demand rather than pre-loaded, connecting more apps does not enlarge the agent's tool surface, and the agent works from each platform's actual schema instead of inventing a field name and writing the wrong thing to a real account.

Apps

Gmail, Outlook, Slack, Microsoft Teams, Discord, Google Calendar, Notion, Linear, Jira, Asana, GitHub, Stripe, QuickBooks, Xero, Shopify, WooCommerce, HubSpot, Salesforce, Pipedrive, Attio, Zendesk, Intercom, Twilio, Airtable, Google Sheets, and many more.

Setup

Remote (recommended), nothing to install:

{
  "mcpServers": {
    "one": {
      "type": "http",
      "url": "https://mcp.withone.ai/mcp"
    }
  }
}

The first connection opens One's consent screen in your browser. There is no API key in the config.

Local, if you prefer stdio:

{
  "mcpServers": {
    "one": {
      "command": "npx",
      "args": ["-y", "@withone/mcp"],
      "env": { "ONE_SECRET": "your-one-api-key" }
    }
  }
}

Scoped access

On the consent screen you choose what the agent gets per connection: full access, read-only, or a specific list of actions. There is also a knowledge-only mode that removes the execute tool entirely, so the agent can read API schemas without firing live requests. Those choices are enforced server-side on every call and reflected back to the agent through each connection's access field, so it knows what it can run without guessing.

Links

Frequently asked questions

What is the One remote MCP server?

The One remote MCP server is a hosted Model Context Protocol endpoint at https://mcp.withone.ai/mcp, so AI assistants can connect to it without installing or running anything locally.

How do I connect to the One MCP server?

Add the endpoint https://mcp.withone.ai/mcp to any MCP-compatible client such as Claude Code, Cursor, or VS Code. The setup snippets on this page configure each client in one step.

Does the One MCP server require authentication?

Yes. One uses OAuth: the first time you connect, your MCP client opens a browser window to sign in and authorize access, then reuses the credentials for future sessions.

Which transport does the One MCP server use?

One exposes a Streamable HTTP endpoint, the transport used by remote MCP servers and supported by all major MCP clients.

Comments