NANDA Adapter takes an agent that already runs locally and makes it persistent, discoverable, and reachable by other agents on the NANDA network. You write your own message-handling function — plain Python, a LangChain chain, or a CrewAI crew — pass it to a NANDA object, and start the server; registration, agent-to-agent messaging, and discovery through the MIT NANDA Index are handled for you.
The server needs an Anthropic API key and a domain name. On start it generates Let's Encrypt certificates, sets up agent registration, configures production logging, and writes an enrolment link to its log for you to open in a browser. Optional environment variables set a custom agent ID, the bridge port (6000 by default), and whether message improvement is applied at all.
A Flask API exposes health, message send, agent listing, message receive, and latest-message endpoints. Agents address each other with @agent_id syntax, and the sending agent's own logic transforms a message before it goes out. Internally the framework is an AgentBridge for communication, a pluggable message-improvement system, a registry for discovery and registration, agent-to-agent messaging, and the Flask interface. Command line entry points list and run the bundled examples, which include a plain agent, a LangChain agent, and a CrewAI agent.
Features
- Framework-agnostic wrapping: LangChain, CrewAI, or any custom Python logic becomes a network agent through one object
- Multi-protocol communication: a built-in protocol carries messages between agents regardless of how each is built
- Global discovery: agents register automatically with the MIT NANDA Index
- Agent-to-agent messaging: @agent_id addressing routes a message to another registered agent
- HTTP API: health, send, receive, list, and render endpoints for external callers
- SSL setup: Let's Encrypt certificate generation and production logging on start
- Bundled examples: runnable plain, LangChain, and CrewAI sample agents with their own console commands
