AI agents today are forced to parse messy HTML, reverse-engineer form flows, and guess permissions — because the web was built for browsers, not bots.
So I built AgentML — append /agents to any resource URL and instead of HTML, you get a structured workspace:
- Who the agent is (identity + role)
- What the current state is
- What actions are available right now
- What's unavailable and why
- Navigation to related resources
Works with FastAPI in 2 lines:
from agentML import AgentML
agent = AgentML(app)
No separate tool server. No rewriting your backend. Your existing OpenAPI spec is enough to get started.
Think of it as MCP but for your existing HTTP API.
GitHub: https://github.com/priyanshu7739410/AgentML
Happy to answer questions — still early (v0.1) but the core convention is solid.
Top comments (0)