WebMCP
MCP for the browser: a web page registers callable tools on document.modelContext so a visiting agent invokes functions instead of clicking pixels.
- name
- WebMCP
- full_name
- Web Model Context Protocol
- layer
- capability
- creator
- W3C Web Machine Learning Community Group (engineers from Google & Microsoft)
- status
- draft (Chrome 146 DevTrial)
- year
- 2026
- one_liner
- MCP for the browser: a web page registers callable tools on document.modelContext so a visiting agent invokes functions instead of clicking pixels.
- spec_url
- https://webmachinelearning.github.io/webmcp/
- snippet
await document.modelContext.registerTool({ name, description, inputSchema, execute })- abbreviation
- WebMCP
- also_known_as
Web Model Context Protocol- canonical_spec_url
- https://webmachinelearning.github.io/webmcp/
- entity_uri
- https://webmachinelearning.github.io/webmcp/
- taxonomy_layer
- capability
- sub_layer
- browser-tooling
- protocol_type
- tooling
- central_problem
- Lets a website expose typed, callable tools to an AI agent running in the browser, so the agent invokes functions directly instead of simulating clicks on the rendered page.
- maintainer
- W3C Web Machine Learning Community Group (editors: Brandon Walderman / Microsoft, Khushal Sagar / Google, Dominic Farolino / Google)
- governance_body
- W3C
- license
- W3C Community Contributor License Agreement (CLA)
- maturity_tag
- experimental
- current_spec_version
- Draft Community Group Report
- spec_date
- 2026-06-15
- launch_date
- — verify-against-primary-at-build ↗ https://webmachinelearning.github.io/webmcp/
- last_verified
- 2026-06-15
- transport
- In-page JavaScript API (navigator.modelContext / document.modelContext)
- core_mechanism
- A page calls navigator.modelContext.registerTool(tool, options) (also exposed as document.modelContext) to register typed tools with name, description, inputSchema, and an execute handler; a browser-resident agent enumerates and invokes those tools. It is a Draft Community Group Report, NOT a W3C Standard and not on the W3C Standards Track.
- discovery_endpoint
- navigator.modelContext / document.modelContext (in-page JS object)
- settlement_type
- —
- adoption_metric
- — verify-against-primary-at-build ↗ https://webmachinelearning.github.io/webmcp/
- notable_adopters
{"value":"Microsoft (editor org)","source":"https://webmachinelearning.github.io/webmcp/"}{"value":"Google (editor org)","source":"https://webmachinelearning.github.io/webmcp/"}- relationships
{"predicate":"standardized_by","target":"w3c","note":"WebMCP -standardized_by-> W3C (the standards body; modeled as a planned governance entity, not yet a record id)."}{"predicate":"extends","target":"mcp","note":"WebMCP brings the MCP tool-calling model to the browser page context."}- ideal_use_case
- Making an interactive web app callable by browser-resident agents (Operator/Comet/Atlas-class) without DOM scraping.
- when_to_use
- When your value is in a rich client-side web UI and you want agents to drive it via typed tools rather than fragile pixel-clicking.
- when_not_to_use
- When you need a server-side, cross-host agent integration (use MCP), or stable production guarantees — WebMCP is an unstable draft and expects breaking changes.
- code_example
- navigator.modelContext.registerTool({ name: 'add_to_cart', description: 'Add a product to the cart', inputSchema: { type: 'object', properties: { sku: { type: 'string' } }, required: ['sku'] }, async execute({ sku }) { return await cart.add(sku); } });
- source
- Status, editors, API surface, draft dated 15 June 2026: https://webmachinelearning.github.io/webmcp/ (Draft Community Group Report; not a W3C Standard). Governance (W3C): brief §L2 + research §2.
- agent_readiness_link
- agent-readiness/webmcp
- layer_legacy
- tool