Why this exists
Agents read the web through markup made for people: menus, consent overlays, tracking scripts, layout. To learn one fact from a WordPress page, an assistant downloads and reasons over tens of thousands of tokens, and the site pays for every pixel. Corsen Context gives the site owner one explicit, bounded, auditable read path instead: four tools with published schemas, canonical URLs, and clean Markdown, sitting beside the human pages without changing them.
Who builds it
Corsen Context is built by Corsen AI, a French company. It started as an open-source MCP and llms.txt layer for websites; the WebMCP registration path, the owner controls, the ten public integrations and the verification tooling were built during The WebMCP Challenge, and the project keeps evolving beyond it.
Principles
- Read-only by contract. The four core tools read; every write-shaped surface is either absent or explicitly human-only.
- Owner control, not agent trust. Post types, paths, tools and purchases are decided by the owner and rendered from one policy table into every channel an agent can read.
- Honest boundaries. Annotations and policy flags are instructions to agents; the server states plainly what it enforces and what it does not.
- Typed results over flattened text. Structured content, explicit errors, and canonical URLs in every answer.
- One contract, verified. A language-neutral manifest, two independent implementations (TypeScript and PHP), parity tests, and a public verifier over the live sites.
What comes next
Adapters for further commerce platforms, per-store governance profiles and owner dashboards are in preparation. Until they ship, they are not claimed here. The current release, changelog and roadmap live in the public repository.