IaGenify SDK 0.3.0: connect agents to MCP servers you choose.

Many useful tools already speak the Model Context Protocol. Before an SDK can use them, developers often need an adapter for each provider, a way to translate tool definitions and a separate connection path for credentials. IaGenify SDK 0.3.0 adds an MCP client so a compatible server can be connected through a common interface.

Connect once, select what is available

The connection identifies the server URL and can include its authentication token. Developers can name the tools they want to make available. If no tool list is supplied, the connection enables no tools by default; choosing an explicit list keeps the integration narrow and understandable.

await api.mcp.connect({
  name: "team-tools",
  url: "https://tools.example.com/mcp",
  token: process.env.TOOLS_TOKEN,
  tools: ["search", "get_record"],
});

The example uses a generic server and environment variable. Application code should keep credentials in its runtime's secret store and should never place a real token in a client bundle, article, log or source repository.

Give an agent a deliberate tool set

Once a connection exists, an agent can be configured to use selected MCP servers. The agent receives the tools associated with those connections alongside its other configured tools. A developer can also retrieve MCP tool definitions and pass them to a custom model-turn loop without adopting the SDK's agent loop.

const agent = api.agents.create({
  runtime: "local",
  machine,
  model: "your-model",
  mcp: ["team-tools"],
});

const definitions = await api.mcp.tools();

The `mcp` option makes the selection visible at the point where the agent is created. Leaving that option out uses the account's connected servers; setting it to an empty list explicitly uses none. That gives application authors a clear way to decide whether an agent should see external tools.

Permissions remain part of the integration

MCP makes tool access easier to connect, but it does not remove the need to think about what those tools can do. The SDK follows the connected server's permissions and the agent's execution mode. An application should choose only the servers and tools needed for a task, then keep confirmation requirements appropriate for actions that change data or affect other people.

This is especially useful when a team already has tools behind an MCP server. Instead of maintaining a separate IaGenify adapter for each service, the developer can use the protocol boundary, keep the integration list explicit and let the existing server define its available operations.

One protocol, two ways to use it

Some applications want IaGenify to run the agent loop and select tools for it. Others already have their own orchestration and need only the MCP tool declarations. Supporting both paths keeps the integration useful without requiring every developer to adopt one application architecture.

Connect the tools your team already has, choose what an agent may see, and keep the integration boundary visible in code.

SDK 0.3.0 brings MCP into the developer surface as a first-class connection type. It makes the protocol easier to use while leaving tool selection and permission decisions with the application that owns the task.