AI tool calling - Nango Docs (original) (raw)

Authorize: Your users connect their accounts via Nango.
Expose tools: You configure which Actions your LLM or agent can call.
Call tools: The agent invokes Nango-hosted Actions via API or MCP.
Observe &amp; optimize: Track results and performance in Nango’s Logs .

Nango makes it simple to authorize APIs and call external tools from any LLM or AI framework.

Use Nango whenever your AI needs to take action in the real world.

You want your agent or LLM workflow to create, update, or fetch data from real APIs.
You want to avoid building your own auth, tool infrastructure, or execution sandbox.
You need reliability, observability, and scalability for tool calls.

Letting an AI agent create or update a CRM opportunity.
Having a chatbot fetch live context from Notion, Linear, or HubSpot.
Running automations triggered by an LLM (e.g., “schedule a meeting” or “create a task”).

Key facts

Framework-agnostic: Works with any agentic or LLM framework (or none at all).
Compatible with OpenAI Agents SDK, Vercel AI SDK, LangChain, LlamaIndex, and more.
Unified API auth: Securely authorize access to third-party APIs via Nango’s Auth system.
You can even surface Connect Links directly in your chat UI for user-driven OAuth.
Tool execution via Actions:
- Tools in Nango are powered by Actions.
- Use prebuilt templates for common tools (e.g. “Create Lead”, “Send Message”).
- Or build your own Actions for full control and reliability.
- Autonomous tool introspection & selection (reference)
MCP-compatible:
- Expose your integrations through Nango’s built-in MCP Server.
- Follow our implement the MCP Server guide.
- Check out the MCP server demo
- Connect to existing official MCP servers (e.g. Notion, Linear) with Nango’s MCP auth.
Secure, scalable, and observable:
- Tool code runs in sandboxed, elastic environments.
- Under 100ms overhead for tool executions — fast enough for human-in-the-loop workflows.
- Logs and metrics to monitor and optimize your tool calls.

Nango works with any framework or SDK that supports tool calling. Here are some example guides to perform tool calling with: