The Model Context Protocol (MCP) is an open protocol that standardizes how applications provide context to large language models (LLMs). By using MCP, you can reduce development time and complexity when building or integrating AI applications and agents.

Remote MCP servers extend AI applications’ capabilities beyond your local environment, giving them access to internet-hosted tools, services, and data sources. Connecting to these servers allows AI assistants to handle complex, multi-step projects with real-time access to external resources. The main benefit of remote MCP servers is their accessibility. Unlike local servers that must be installed and configured on individual devices, remote servers can be reached from any MCP client with an internet connection. Remote MCP servers allow AI21 Maestro to connect with internet-hosted tools, services, and data sources in real time. This makes them ideal for complex, multi-step workflows that require secure server-side processing and authentication.

When you define an MCP tool in the tools parameter, AI21 Maestro acts as an MCP client. It authenticates with the server, retrieves the list of available tools defined there, and can invoke them during a session to fetch data or perform actions.
AI21 Maestro supports only remote MCP servers, not local ones.

​
When to use MCP

Use MCP when you need:
  • Dynamic discovery of available tools from a remote server.
  • Access to multiple related tools (for example, a Jira MCP server exposing several Jira-related functions).
  • Secure, server-side integration with authentication.
  • Reusable integrations that can be shared across multiple sessions or projects.

​
How to Register an MCP Server

​
Required fields:

FieldDescription
typeMust be "mcp"
server_labelUnique name for the MCP server within the session
server_urlFully qualified URL of the MCP server endpoint
headers (optional)Authentication headers (e.g., bearer tokens)
allowed_tools (optional)List of tool names to import; if omitted, all tools will be loaded

​
JSON Example

{
  "type": "mcp",
  "server_label": "Jira",
  "server_url": "https://app.hubspot.com/mcp/v1/http",
  "headers": {
    "Authorization": "Bearer: <token>"
  },
  "allowed_tools": [
    "getJiraIssue",
    "searchJiraIssuesUsingJql",
    "getVisibleJiraProjects"
  ]
}

​
Authentication

MCP supports common authentication schemes used in HTTP-based APIs. You can include credentials via headers when defining the tool.

​
Supported Methods:

  • Bearer tokens in the Authorization header
  • API keys in custom headers
  • OAuth tokens or session-based headers

​
Best Practices

  • Use descriptive tool names to make them easy for agents to understand and select.
  • Limit tools with allowed_tools if your server exposes many capabilities.
  • Write complete descriptions for functions and parameters—these are used by the model to decide when and how to invoke them.
  • Secure your endpoints using proper authentication and avoid exposing unnecessary tools.

​
Next steps

Build an Employee Server with Fast MCP.