django-agentweb

Make your Django site a first-class citizen of the agentic web.

django-agentweb is a reusable Django app that helps AI agents (and the LLMs behind them) discover, understand, and act on your site — across five complementary domains. Every domain is opt-in per site and off by default, so you expose only what you choose.

Status: alpha. Core domains (llms.txt, JSON-LD profiles, discovery / ai-catalog + MCP server-card, browser WebMCP, CAP Lite) are implemented and opt-in per site. Booking adapters and extras (Web Bot Auth, full SDF) follow. See the roadmap.

The five agent-web domains

Why

Search and assistants increasingly mediate discovery through agents. For an independent hotel, that means letting an agent check availability, simulate a booking price, and start a direct reservation — and reducing dependence on OTAs (Booking.com / Expedia). The same building blocks benefit any content or commerce site.

This package is intentionally generic and externally adoptable: it implements all five domains even where a given deployment won't use them, to maximise reuse and external visibility.

Install

Until the package is published on PyPI, install from VCS or an editable checkout:

pip install "django-agentweb @ git+https://github.com/DLRSP/django-agentweb.git"
# or: pip install -e ".[testing]"

Extras once installed:

pip install "django-agentweb[webmcp]"  # + server-side tool proxy
pip install "django-agentweb[all]"     # everything

Requires django.contrib.sites.

Quickstart

# settings.py
INSTALLED_APPS = [
    # ...
    "django.contrib.sites",
    "agentweb",
]

AGENTWEB = {
    "LLMS": {"ENABLED": True},
    "JSONLD": {"ENABLED": True},
    "DISCOVERY": {"ENABLED": True},
    "WEBMCP": {"ENABLED": False},
    "COMMERCE": {"ENABLED": False},
    "SDF": {"ENABLED": False},
}

# urls.py
from django.urls import include, path

urlpatterns = [
    # ...
    path("", include("agentweb.urls")),
]

Only the domains you enable register any URLs or expose any data.

Security

Agent-facing surfaces have a distinct threat model (prompt/output injection, data leakage, unsafe tool calls). Transactional tools always require human-in-the-loop. See SECURITY.md.

Roadmap

Readability → structured data → discovery → WebMCP → commerce/booking → SDF, followed by per-site rollout. See the package docs for the current domain status.

License

MIT — see LICENSE.