WebMCP implementation

Goal

Give an existing web app a small, coherent agent-facing interface to its real capabilities. Agents should be able to discover useful actions, provide validated inputs, and receive verifiable results while people retain the same UI, permissions, and control. Implement the integration, not just a proposal, unless the user requests research or planning only.

Work with the application's architecture. Do not build a second application or duplicate its business logic.

Read selectively

Read compatibility before selecting APIs. Read implementation when designing the adapter and tool contracts, then verification before testing. Use the planning template for the implementation record. The optional browser probe inspects capabilities without invoking or registering tools.

These references contain a dated research snapshot and recommended engineering policy. They are not a substitute for checking the target runtime.

1. Inspect the repository and establish scope

Read applicable AGENTS.md, project instructions, dependency manifests, routes, authentication, state management, validation, domain services, and test configuration. Respect the existing package manager and UI/design system. Inspect the working tree and preserve unrelated changes.

Identify the real human workflows and trace their implementation end to end. Note the browser/client boundary and where authorization is enforced. For a monorepo, work in the relevant web application; do not add a repo-wide framework merely to expose one app.