Skip to main content
WebMCP is a Draft Community Group Report, not a completed W3C standard. It defines the direction of the browser API, but several details remain a work in progress. The imperative draft now defines cross-document exposure, discovery, and the tools Permissions Policy. Declarative attributes, testing surfaces, browser-agent mediation, cancellation, and output-schema behavior may continue to change. Chrome offers WebMCP through an origin trial beginning in Chrome 149. Local development uses the dedicated chrome://flags/#enable-webmcp-testing flag. Track the Chrome WebMCP overview for current availability. Implementations should track the current Community Group draft, keep domain operations separate from the browser registration adapter, use feature detection, and test both native and polyfill paths. MCP-B extensions must remain clearly distinguished from the standard surface. The current proposal defines document.modelContext. Navigator-based model context APIs are not part of that surface; MCP-B retains them only as compatibility shims. Adapter boundaries reduce migration cost. See Try the native Chrome preview and Test native and polyfill.

Upstream sources to track

Use first-party sources for compatibility decisions: MCP-B follows those boundaries by making document.modelContext the primary surface, using getTools() for discovery, feature-detecting Chrome’s optional executeTool() extension, using AbortSignal for cleanup, and keeping legacy helpers as compatibility APIs. outputSchema is not part of the current W3C/Chrome WebMCP tool dictionary. MCP-B keeps it as helper metadata for TypeScript inference, structured MCP responses, and package/runtime adapters. Do not rely on native browser WebMCP to validate tool output against it.