From API to
MCP server.

Templates cover 28 public APIs, from OpenAI, Slack, and Jira to PubMed and ClinicalTrials.gov. Choosing one imports the API’s published contract with a starter set of tools selected, and tells you how to get its credential. A few need a setting first, such as your Zendesk or Jira address; the template says so.

Tool names and descriptions guide the model when deciding whether to call an endpoint. Keep tools narrow, remove destructive operations unless required, and state side effects explicitly. If AI descriptions are on, the workspace can suggest one to start from.

A toolkit serves tools from several of your servers together, so an agent connects and signs in once. Open Toolkits in the workspace, choose tools from your servers’ published tools, and create it. Each tool is named after its server, such as zendesk_ListTickets, and any name can be changed.

Every call still runs through its own server, with that server’s credentials, limits, and call log. A toolkit has its own tokens and OAuth sign-in, and connects like a server:

Upstream credentials are encrypted at rest. A server can store one shared credential, or use each person’s own, which each member sets for themselves. For APIs that use OAuth 2, register an app with the provider and connect through its consent screen; tokens refresh automatically. Environments point the same tools at another deployment, such as staging, with its own credential if you like.

Copy the MCP endpoint from a server’s or toolkit’s Overview and add it as a remote MCP server in your client. The client reads the OAuth discovery metadata, opens the AlltoMCP authorization flow, and receives scoped access after you approve it. Cursor and VS Code install with one click from the same page, and Claude Code with one command.

Each server and toolkit can have multiple named tokens. A token can be limited to read-only tools or to chosen tools, and set to expire. Its secret is shown once when created or rotated; only its hash and display prefix are retained.

Every account has a personal workspace. Create a team workspace to share servers and toolkits, and invite people by link with one of four roles. Every change is recorded in the workspace’s audit log, with who made it and how.

Scripts and CI can manage servers, tools, and tokens through the same API the workspace uses. Create a personal API key under Account → API keys; it acts as you, with your role in each workspace. Read-only keys can only read, keys can expire, and changes made with one show in the audit log.

Every workspace starts on the free plan. Plans set how many MCP servers and toolkits a workspace can have, how many tool calls it can make each calendar month (UTC), and how many members a team workspace can have, counting pending invites. Every feature is on every plan.

At a limit, creating another server or toolkit, or inviting another member, is refused with a message naming the limit, and tool calls past the monthly allowance return an error the agent can relay. A workspace’s plan and usage are under Billing in the workspace.