Proposal: An `AIDLoader` or `from_domain()` constructor for dynamic Tool creation

Talking Shop
1

Hey LangChain forum,

One of the most powerful features of LangChain is its vast library of integrations and tools, which abstracts away the pain of connecting to various APIs. However, setting up a new custom tool, like with OpenAPITool, still often requires the developer to find the specific spec URL and manually configure the connection.

I’ve been thinking about how we could simplify this down to the absolute minimum: just a domain name.

I’ve recently launched a simple, open standard called Agent Interface Discovery (AID) that enables this. It’s essentially an “MX record for AI agents,” using a single DNS TXT record to make an agent’s interface discoverable.

For example, a service could publish a record like this:

_agent.weather-api.com. IN TXT "v=aid1;p=openapi;uri=https://api.weather-api.com/v1/spec.json"

The “LangChain-Native” Experience

With this in place, we could introduce a new loader or tool constructor that would feel incredibly intuitive to a LangChain developer. Imagine this workflow:

from langchain_community.tools import load_tool_from_aid # or a new AIDTool class

# Instead of hunting for a spec URL, the developer just uses the domain.
weather_tool = load_tool_from_aid("weather-api.com")

# The loader would:
# 1. Perform the AID DNS lookup for "weather-api.com"
# 2. Find the URI and the protocol ("openapi")
# 3. Intelligently call the appropriate existing LangChain loader (e.g., OpenAPITool.from_spec_url)
# 4. Return a fully instantiated and usable tool.

# Now the tool is ready to be used in an agent
# >>> weather_tool.run("What's the weather in San Francisco?")

Why this fits LangChain’s goals:

  • Radically Simplifies Developer Experience: It reduces the tool connection process to its logical minimum.
  • Enables Dynamic & Autonomous Agents: An agent could be given a list of domains and dynamically load the tools it needs for a task, without those tools being pre-configured. This is a step towards more autonomous systems.
  • Massively Extensible: Any service that publishes an AID record instantly becomes compatible with LangChain through a single, generic loader.

This feels like a natural extension of LangChain’s philosophy of making complex compositions simple. It’s a small, decentralized standard that could have a huge impact on the developer workflow.

The spec is final, with libraries available. Would love to get the community’s thoughts on this and discuss what the best integration path might look like.

You can find the full spec, docs, and a live workbench here: aid.agentcommunity.org