WebMCP: Web Model Context Protocol Performance PlanA performance plan for the WebMCP specification effort within the W3C Web Machine Learning Community Group, specifying goals, objectives, stakeholder roles, and performance indicators for defining, standardizing, securing, and fostering adoption of a browser-native API (navigator.modelContext) enabling AI agent interaction with websites without backend infrastructure.Source: "WebMCP: Everything You Need to Know," prepared by Claude (Anthropic) and prompted by Paola Di Maio, W3C AI-KR Community Group, February 2026. Part 1 StratML extraction: Claude / Owen Ambur, February 17, 2026. Part 2 performance indicators and stakeholder roles: proposed, not formally adopted by the W3C WebML CG. All TargetResult values are proposed planning targets. Full source document: https://starborn.github.io/webmcp/webmcp-complete-guide.html W3C Web Machine Learning Community GroupW3C WebML CGuuid-org-w3c-webml-cgThe W3C Community Group incubating the WebMCP draft specification, led by engineers from Google and Microsoft. Responsible for producing the Draft Community Group Report and advancing the specification toward the W3C Standards Track.GooglePrimary engineering contributor to the WebMCP specification; building experimental browser-side agent support into Chrome and leading API design within the WebML CG.Specification Author and Browser ImplementerDrafts core API design, authors WebIDL, and commits to experimental Chrome implementation behind a flag.PerformerBeneficiaryMicrosoftContributing engineering organization to the WebMCP specification effort; brings browser engine and enterprise deployment perspective.Specification ContributorReviews and contributes to WebIDL and security model sections; represents Edge browser implementation interests.PerformerBeneficiaryWorld Wide Web ConsortiumHost standards body providing Community Group infrastructure, IP framework, and the eventual Working Group process.Standards Process HostProvides the governance structure, IP commitments framework, and publication infrastructure for advancing WebMCP to a full W3C Recommendation.PerformerBeneficiaryA web in which any website can be an AI-friendly service — exposing its capabilities directly to AI agents through the browser, without requiring backend server infrastructure — enabling seamless, secure, user-present human-AI-web interaction at internet scale.uuid-vision-webmcpTo define, standardize, and foster adoption of a browser-native API (navigator.modelContext) that allows web developers to register machine-callable tools within their existing JavaScript code, enabling AI agents to discover and invoke website capabilities without screen scraping or separate backend MCP server infrastructure, while preserving user consent and the web's established origin-based security model.uuid-mission-webmcpSecurity by InheritanceWebMCP builds on the web's established security primitives rather than inventing new ones. Origin isolation, HTTPS-only (SecureContext), and same-origin enforcement are the foundation. Trust is earned through the browser's existing model, not invented from scratch.User-Present DesignWebMCP is designed for interactions in which the human is present at the browser and can be asked for confirmation before consequential actions are taken. The requestUserInteraction() mechanism embodies the principle that autonomous action without human awareness is not the default.Open Standards ProcessThe specification is developed openly through W3C Community Group infrastructure, with public participation invited before design decisions are locked in.Progressive AdoptionWebMCP is designed for incremental adoption using existing JavaScript without new backend infrastructure. Polyfills (MCP-B) enable adoption before native browser support arrives.API DefinitionDefine a complete, minimal, and implementable browser API (navigator.modelContext) through which web developers can register, update, and remove machine-callable tools, and through which AI agents can discover and invoke those tools within the browser's security boundary._a3f49049-b486-4b07-a396-923f5d88c0311Web DevelopersDevelopers who register tools in existing JavaScript to expose site capabilities to AI agents via WebMCP.API Implementer and AdopterImplements tool registrations using navigator.modelContext on their websites; primary beneficiary of a stable, well-documented API.PerformerBeneficiaryAI Platform ProvidersCompanies such as Anthropic, OpenAI, and Google whose agents (Claude, ChatGPT, Gemini) call tools registered via WebMCP.Agent ImplementerIntegrates WebMCP tool discovery and invocation into AI agent products; benefits from a standardized interface requiring no per-site custom integration.PerformerBeneficiaryBrowser VendorsCompanies implementing WebMCP natively in browser engines, beginning with Chrome (Google).Platform ImplementerImplements navigator.modelContext in browser engines and enforces SecureContext and origin isolation requirements.PerformerCore API surface: navigator.modelContext (entry point); provideContext(options) — register full tool set, replacing prior set; clearContext() — remove all tools; registerTool(tool) — add one tool dynamically; unregisterTool(name) — remove one tool by name. Tool object fields: name, description, inputSchema (JSON Schema), execute (callback), annotations (readOnlyHint). Agent identity surface: ModelContextClient with requestUserInteraction(callback) for human-in-loop consent during execution.navigator.modelContext InterfaceSpecify and implement the navigator.modelContext interface as the single, stable entry point for all WebMCP functionality, accessible in SecureContext (HTTPS) browser windows only, returning the same object on every access (SameObject semantics)._afaa9592-b316-44b0-ab1c-bac427912ee31.1Browser VendorsImplementers of the navigator.modelContext interface in browser engine code.Interface ImplementerImplements the SameObject navigator.modelContext interface in browser engine; enforces SecureContext restriction.Performer1.1.aInterface specification completenessPublished WebIDL interface definition (complete / incomplete)pi-1-1-aPLACEHOLDER_112027-06-30StatusCompleteA single, complete WebIDL interface definition for navigator.modelContext published in the W3C Draft specification, covering SameObject semantics and SecureContext restriction. ^^ 1 = Complete | 2 = Incomplete[To be determined]1.1.bConformance tests for interface entry pointNumber of Web Platform Tests (WPT) test casespi-1-1-bPLACEHOLDER_22027-12-31StatusTested3Minimum 3 WPT test cases: (1) navigator.modelContext defined in SecureContext; (2) SameObject identity assertion; (3) undefined in non-SecureContext (HTTP).[To be determined]Tool Registration MethodsDefine and stabilize the four tool lifecycle methods — provideContext, clearContext, registerTool, unregisterTool — with clear semantics for replacement vs. additive registration._73f50c75-bc38-4639-96fd-0c52104859de1.2Web DevelopersPrimary users of the tool registration lifecycle methods.API ConsumerCalls provideContext, registerTool, unregisterTool, and clearContext to manage tool availability throughout the page lifecycle.PerformerBeneficiary1.2.aTool lifecycle methods specified with normative semanticsNumber of methods with complete specification and conformance testspi-1-2-aPLACEHOLDER_32027-06-30StatusSpecified4All four methods (provideContext, clearContext, registerTool, unregisterTool) specified with normative prose, edge-case handling, and at least one conformance test each.[To be determined]1.2.bDeveloper ergonomics and predictability of registration APIAbsence of known ambiguity bugs filed against lifecycle semantics (yes / no)pi-1-2-bPLACEHOLDER_402028-06-30StatusNoneNo unresolved issues in the W3C spec repository regarding ambiguous behavior of provideContext vs. registerTool replacement semantics.[To be determined]Tool Object SchemaSpecify the ModelContextTool dictionary — name, description, inputSchema (JSON Schema), execute callback, and annotations — so that tool definitions are self-describing, machine-validatable, and legible to AI agents without additional documentation._d2d9e872-8dac-47cf-91c4-79ce06abddc21.3AI Platform ProvidersConsume tool definitions to decide whether and how to call each tool; require self-describing schemas.Tool Definition ConsumerReads name, description, and inputSchema to determine tool purpose, required inputs, and safety (readOnlyHint).Beneficiary1.3.aModelContextTool dictionary fields specifiedNumber of fields with normative definition and JSON Schema examplepi-1-3-aPLACEHOLDER_52027-06-30StatusSpecified5All five fields (name, description, inputSchema, execute, annotations) defined normatively in WebIDL with at least one worked JSON Schema example in the specification.[To be determined]1.3.bAgent legibility of tool descriptions without documentation lookupConformance to description-field guidance (compliant / non-compliant)pi-1-3-bPLACEHOLDER_612027-12-31StatusBest-practice authoring guide for description field publishedA normative or informative best-practice guide for writing agent-legible tool descriptions is published as part of or alongside the specification. ^^ 1 = Published | 0 = Unpublished[To be determined]Human-in-Loop MechanismSpecify requestUserInteraction(callback) on ModelContextClient so that tool execute functions can pause mid-execution to present a confirmation prompt to the user before completing consequential actions, with the agent waiting on the Promise resolution before proceeding._c6b7e764-1730-4782-a8e6-32ee3990f8951.4End UsersHumans present at the browser who receive confirmation prompts via requestUserInteraction before consequential actions are taken on their behalf.Consent ProviderReviews and approves or rejects agent actions mid-execution via browser-presented confirmation prompts.BeneficiaryWeb DevelopersImplement requestUserInteraction calls within execute functions for tools that make consequential changes.Consent ImplementerInserts requestUserInteraction calls at appropriate points within tool execute functions.Performer1.4.aSpecification completeness of requestUserInteraction mechanismNormative specification published (yes / no)pi-1-4-aPLACEHOLDER_712027-06-30StatusPublished with normative prose and at least one code exampleThe ModelContextClient interface and requestUserInteraction method are specified normatively with a worked code example showing a mid-execution purchase confirmation flow. ^^ 1 = yes | 0 = no[To be determined]1.4.bConformance test coverage for human-in-loop mechanismNumber of WPT test cases covering requestUserInteraction behaviorpi-1-4-bPLACEHOLDER_82028-06-30StatusTested4Minimum 4 WPT test cases: (1) callback invoked before execute resolves; (2) execute cancelled on user rejection; (3) execute proceeds on user approval; (4) Promise chain timing.[To be determined]Security ModelEnsure that WebMCP tool execution operates within the web's established security boundaries — origin isolation, encrypted transport, and user-present consent — so that adoption does not introduce new attack surfaces beyond those already present in the browser platform._c5b3259c-1930-4745-b7a8-1aad5b5222d72End UsersHumans present at the browser whose data, actions, and consent are protected by the WebMCP security model.Security BeneficiaryPrimary beneficiary of origin isolation, SecureContext requirements, and consent gap resolution; trusts that AI agents calling WebMCP tools cannot exfiltrate cross-origin data.BeneficiaryBrowser VendorsImplementers responsible for enforcing SecureContext and origin isolation at the platform level.Security EnforcerImplements platform-level enforcement of origin isolation and SecureContext requirements; primary performer of security model commitments in the specification.PerformerBeneficiaryThree security layers: (1) Origin isolation — tools registered on one origin cannot access another origin's data; (2) SecureContext requirement — WebMCP only available on HTTPS; (3) User-present model — requestUserInteraction provides per-action consent. Security inherits from existing web primitives rather than introducing new mechanisms.Origin IsolationEnforce that tools registered on a given origin operate exclusively within that origin's security boundary, preventing cross-origin data access or action execution by agents calling WebMCP tools._1bd4e3a9-7009-4ef2-825e-11bc08dd522a2.1Browser VendorsEnforce origin isolation at the platform level for WebMCP tool registrations.Origin Boundary EnforcerEnsures that tools registered at one origin cannot be discovered or invoked by agents operating in a different origin's context.Performer2.1.aKnown cross-origin data access vulnerabilities in WebMCP implementationNumber of open cross-origin vulnerability issues in browser bug trackerpi-2-1-aPLACEHOLDER_92028-12-31StatusConfirmed0Zero open, confirmed cross-origin access vulnerabilities attributable to WebMCP tool registration or invocation in any participating browser's bug tracker.[To be determined]2.1.bConformance tests for cross-origin tool isolationNumber of WPT test cases asserting cross-origin tool invisibilitypi-2-1-bPLACEHOLDER_102027-12-31StatusTested2Minimum 2 WPT test cases: (1) tools registered at origin A are not discoverable from an iframe at origin B; (2) agent invocation from origin B cannot call tools registered at origin A.[To be determined]SecureContext EnforcementRequire HTTPS (SecureContext) for all WebMCP functionality, preventing tool registration and agent tool calls on unencrypted HTTP pages to eliminate eavesdropping on tool invocations and responses._ce21dc59-a213-490c-97a4-a21faf7b50142.2Web DevelopersRequired to serve WebMCP-enabled pages over HTTPS; this is the compliance obligation of site operators.HTTPS OperatorEnsures that any page registering WebMCP tools is served over HTTPS; bears responsibility for maintaining valid TLS certificates.PerformerBeneficiary2.2.aSecureContext requirement normatively specified and testedSpecification normative requirement present and WPT test published (yes / no)pi-2-2-aPLACEHOLDER_1112027-06-30StatusNormative SecureContext requirement in spec with accompanying WPT testThe specification normatively states that navigator.modelContext is only available in SecureContext, and at least one WPT test confirms it is undefined (or throws) on an HTTP origin. ^^ 1 = Secure | 2 = Unsecured[To be determined]Consent Gap ResolutionAddress the open design question of whether tool discovery (not just execution) should require explicit user permission, analogous to the camera/microphone/location permission prompt model, or whether the AI client layer is the appropriate place to manage agent access to tool manifests. Resolve the intended architecture and document it in the specification._ddd57691-3567-4876-b6fa-20d5f3d6f00b2.3End UsersHumans whose awareness of and consent to tool discovery is the subject of this open design question.Consent SubjectThe person whose permission to expose tools to AI agents is — or should be — sought; the central stakeholder in the consent gap debate.BeneficiaryAI Platform ProvidersMay be the appropriate layer for managing tool discovery consent if the specification delegates this to the AI client rather than the browser.Consent Architecture Decision-MakerParticipates in resolving whether consent for tool discovery belongs at the browser platform level or the AI agent client level.PerformerBeneficiaryCurrently, any website can silently register tools discoverable by connected agents with no browser-level permission prompt. The per-action requestUserInteraction mechanism exists, but pre-discovery consent does not. This is a key open question for the specification team as of February 2026. Resolution of this question affects the trust model for all WebMCP-enabled deployments.2.3.aArchitectural decision on tool discovery consent documented in specificationDocumented architectural decision with rationale (resolved / unresolved)pi-2-3-aPLACEHOLDER_1212027-06-30StatusResolved — architectural decision documented with rationaleThe specification explicitly states whether tool discovery requires browser-level user permission (like camera/microphone) or whether this responsibility is delegated to the AI client layer, with rationale for the chosen architecture. ^^ 1 = Resolved | 0 = Unresolved[To be determined]2.3.bStakeholder feedback on consent architecture solicited and reviewedNumber of formal public comments on consent gap received and responded topi-2-3-bPLACEHOLDER_132027-03-31StatusAddressed10At least 10 formal public comments on the consent gap issue solicited, received, and addressed in the specification issues tracker before the architectural decision is finalized.[To be determined]Ecosystem ToolingDevelop and publish open-source tooling that supports quality, documentation, validation, and discoverability of MCP servers and WebMCP browser tools, lowering implementation barriers and establishing consistent standards for the ecosystem._c11e0739-851a-4fdd-a8e2-61e362423aec3StarbornOpen-source contributor responsible for the MCP Server Generator, MCP Server Validator, MCP Model Card Generator, MCP Model Card Specification v1.0, and WebMCP Model Card Generator tools. (GitHub: Starborn)Primary Tool DeveloperDesigns, builds, and maintains all five ecosystem quality tools; responds to community issues and publishes releases.PerformerBeneficiaryMCP Server DevelopersDevelopers building backend MCP servers who benefit from generator, validator, and model card tooling.Tooling ConsumerUses MCP Server Generator to scaffold servers, Validator to check security, and Model Card Generator to document them.BeneficiaryWebMCP Tool AuthorsWeb developers building browser-side WebMCP tools who benefit from the WebMCP-specific model card generator and documentation standards.Browser-Side Tooling ConsumerUses WebMCP Model Card Generator to produce standardized documentation for browser-side tool registrations.BeneficiaryFive tools form the quality pipeline: (1) MCP Server Generator — scaffold production-ready server code; (2) MCP Server Validator — static analysis scoring Critical to Excellent; (3) MCP Model Card Generator — machine and human-readable documentation; (4) MCP Model Card Specification v1.0 — six section formal standard; (5) WebMCP Model Card Generator — twelve section browser-specific documentation. Tools 1-4 target backend MCP; tool 5 targets browser-side WebMCP.MCP Server GeneratorProvide an open-source scaffold generator that produces production-ready MCP server code from a natural-language description, reducing implementation time and promoting consistent server structure. (github.com/Starborn/MCP-Server-Generator)_42fa4d5e-b7dc-48e6-ac4b-64038f5bfcbb3.1StarbornDeveloper and maintainer of the MCP Server Generator tool.Tool DeveloperBuilds, releases, and maintains the scaffold generator; processes community contributions and bug reports.Performer3.1.aStable releases of MCP Server Generator publishedNumber of tagged GitHub releasespi-3-1-aPLACEHOLDER_142027-12-31StatusStable2Minimum 2 stable tagged releases of MCP Server Generator published on GitHub, each producing syntactically valid, runnable MCP server scaffold code from a natural-language description.[To be determined]3.1.bCommunity adoption of MCP Server GeneratorNumber of GitHub starspi-3-1-bPLACEHOLDER_152028-12-31StatusAdopted100At least 100 GitHub stars on the MCP Server Generator repository, indicating meaningful community discovery and adoption.[To be determined]MCP Server ValidatorProvide static analysis tooling that scores MCP server code for security vulnerabilities, naming compliance, and known attack patterns (including tool poisoning) without executing the code, giving developers actionable remediation guidance before deployment. (github.com/Starborn/MCP-Server-Validator)_b897f25c-2d3f-40c7-9a0a-f0e67545d6623.2MCP Server DevelopersRun the validator against their server code before deployment to surface security and quality issues.Validator UserSubmits MCP server code to the validator; receives security score and remediation guidance; acts on findings before deployment.PerformerBeneficiary3.2.aVulnerability and attack pattern categories detected by the validatorNumber of distinct vulnerability/attack pattern detection rulespi-3-2-aPLACEHOLDER_162027-12-31StatusDetected10Validator detects at least 10 distinct rule categories including: hardcoded credentials, missing input validation, tool poisoning patterns, insecure transport, and naming convention violations.[To be determined]3.2.bScoring rubric completeness and actionabilityFive-tier scoring rubric published with per-tier remediation guidance (published / not published)pi-3-2-bPLACEHOLDER_172027-06-30StatusPublished — Critical / Poor / Fair / Good / Excellent tiers with remediation stepsAll five scoring tiers (Critical below 25%, Poor 25-49%, Fair 50-74%, Good 75-89%, Excellent 90-100%) documented with specific remediation steps available to developers.[To be determined]MCP Model Card SpecificationPublish and maintain a formal six-section model card specification (server identity, tool documentation, operational characteristics, security profile, deployment context, evaluation results) as a shared documentation standard for the MCP ecosystem. (starborn.github.io/MCP-Model-Card-Generator/)_7e015c4b-6621-42ef-9fb3-2666fb93ecfc3.3StarbornAuthor and publisher of the MCP Model Card Specification v1.0.Specification AuthorDefines, publishes, and maintains the six-section model card standard; accepts community pull requests and issue reports.PerformerBeneficiary3.3.aFormal sections in the MCP Model Card SpecificationNumber of normatively defined sectionspi-3-3-aPLACEHOLDER_182026-12-31StatusPublished6All six sections (server identity, tool documentation, operational characteristics, security profile, deployment context, evaluation results) normatively defined and published at starborn.github.io/MCP-Model-Card-Generator/.[To be determined]WebMCP Model Card GeneratorProvide a twelve-section documentation generator specific to browser-side WebMCP tools, covering navigator.modelContext API modes, origin-based security, user interaction patterns, and browser compatibility testing — addressing the distinct concerns of frontend tool authors versus backend server developers. (starborn.github.io/webmcp/)_11940388-4f56-432d-8b23-811afb5d20633.4WebMCP Tool AuthorsUse the WebMCP Model Card Generator to produce standardized twelve-section documentation for their browser-side tool registrations.Documentation Generator UserInputs tool metadata and receives a structured twelve-section model card in JSON (machine-readable) and Markdown (human-readable) formats.Beneficiary3.4.aBrowser-specific documentation sections in WebMCP Model Card GeneratorNumber of sections covered in the generator outputpi-3-4-aPLACEHOLDER_192026-12-31StatusPublished12All twelve sections covering browser-specific concerns (including navigator.modelContext API modes, origin-based security, user interaction patterns, and browser compatibility testing) implemented and published at starborn.github.io/webmcp/.[To be determined]3.4.bTime from new WebMCP spec publication to corresponding generator updateDays between WebMCP spec change and corresponding generator releasepi-3-4-bPLACEHOLDER_20302028-12-31ResponsivenessGenerator updated within 30 days of any normative WebMCP spec changeThe WebMCP Model Card Generator is updated within 30 calendar days whenever a normative change to the WebMCP specification affects model card content, as demonstrated by release notes referencing the spec change.[To be determined]Standards AdvancementAdvance WebMCP from Draft Community Group Report through the W3C standards process toward a full W3C Recommendation, while enabling early experimental browser implementation and ecosystem adoption in parallel with the formal process._05071cd1-887a-48ee-94b8-68dd558869304W3C Web Machine Learning Community GroupIncubating the spec and producing the Draft Community Group Report that initiates the standards track.Specification IncubatorProduces and publishes the Draft Community Group Report; solicits feedback; develops Working Group charter proposal.PerformerBeneficiaryGoogle ChromeBrowser implementer expected to ship experimental WebMCP support behind a flag in Chrome ahead of full standardization.Early ImplementerShips experimental navigator.modelContext behind a Chrome flag; provides real-world implementation feedback to the specification process.PerformerBeneficiaryStandards ContributorsTechnical contributors — tool authors, AI platform engineers, web developers, and security researchers — participating in the specification process through W3C channels.Specification Reviewer and ContributorReviews draft specification text, files issues, submits pull requests, and participates in Community Group calls and Working Group processes.PerformerBeneficiaryCurrent status as of February 2026: Draft Community Group Report. Not yet on W3C Standards Track; not yet a W3C Recommendation. Typical path: Community Group Report → Working Group charter → Working Draft → Candidate Recommendation → W3C Recommendation. Full standardization typically takes years. Chrome experimental (flag) implementation may arrive significantly sooner. MCP-B polyfill provides interim implementation capability.Working Group CharterDevelop a Working Group charter proposal to bring WebMCP onto the formal W3C Standards Track, transitioning it from Community Group incubation to an official W3C deliverable with IP commitments from participating organizations._be02b412-d5b6-400b-9e84-8e2766066e984.1W3C Web Machine Learning Community GroupResponsible for drafting and submitting the Working Group charter proposal to the W3C Advisory Committee.Charter AuthorDrafts scope, deliverables, timeline, and participation requirements for the proposed Working Group charter.Performer4.1.aWorking Group charter proposal milestoneCharter proposal submitted to W3C Advisory Committee (yes / no)pi-4-1-aPLACEHOLDER_2112027-12-31StatusCharter proposal submitted to W3C Advisory Committee for reviewA formal Working Group charter proposal for WebMCP has been drafted, reviewed by Community Group participants, and submitted to the W3C Advisory Committee for its consideration by end of 2027. ^^ 1 = Submitted | 0 = Unsubmitted[To be determined]Chrome Experimental ImplementationShip an experimental implementation of navigator.modelContext behind a browser flag in Chrome, enabling real-world developer testing and feedback to inform specification refinement before the standard is finalized._fe6f80c3-73db-483f-a502-87aa56d411b44.2Google ChromeBrowser engineering team responsible for the experimental implementation behind a Chrome flag.Experimental ImplementerImplements navigator.modelContext in Chromium source; ships behind an intent-to-experiment flag; collects developer feedback via chromium issue tracker.PerformerBeneficiary4.2.aExperimental Chrome implementation shipped behind flagImplementation available in Chrome Canary / Dev channel (yes / no)pi-4-2-aPLACEHOLDER_2212027-06-30Statusnavigator.modelContext available in Chrome Canary behind flagnavigator.modelContext is available behind a Chrome experiment flag in Chrome Canary or Dev channel, enabling web developers to test tool registration and agent invocation against a real browser implementation. ^^ 1 = Implemented | 2 = Unimplemented[To be determined]4.2.bDeveloper feedback issues filed against experimental implementationNumber of developer-filed issues in Chromium bug tracker addressed before specification freezepi-4-2-bPLACEHOLDER_232028-06-30StatusAddressed20At least 20 developer-filed implementation feedback issues reviewed and addressed (resolved, wontfix, or deferred with rationale) before the specification advances to Candidate Recommendation.[To be determined]Polyfill Bridge (MCP-B)Maintain MCP-B as a community-supported polyfill implementing navigator.modelContext via browser extension and JavaScript library, translating between WebMCP's browser-native format and MCP's JSON-RPC wire protocol, so developers and AI clients can adopt WebMCP today without waiting for native browser support. (WebMCP-org on GitHub)_272eae70-43ed-4123-9563-de16c42d62054.3WebMCP-org CommunityThe open-source community maintaining the MCP-B polyfill on GitHub under the WebMCP-org organization.Polyfill MaintainerMaintains the MCP-B browser extension and JavaScript library; keeps the polyfill synchronized with the evolving WebMCP spec.PerformerBeneficiaryAI Platform ProvidersUse MCP-B to connect their existing MCP clients to browser-side WebMCP tools before native browser support arrives.Polyfill ConsumerIntegrates MCP-B transport into their AI agent clients to enable calling WebMCP tools through the existing MCP wire protocol.Beneficiary4.3.aWebMCP API surface covered by MCP-B polyfillPercentage of navigator.modelContext methods implemented in MCP-Bpi-4-3-aPLACEHOLDER_242027-12-31StatusImplemented100MCP-B implements 100% of the navigator.modelContext API surface: provideContext, clearContext, registerTool, unregisterTool, and ModelContextClient.requestUserInteraction.[To be determined]Public Feedback IntegrationSolicit and incorporate structured feedback from the broader web development community, AI platform providers, security researchers, and user advocates — particularly on the consent gap and the intended architecture for agent-client-level vs. browser-level access control — before the specification advances beyond Draft status._5e7c6e97-eee7-4781-8ac7-f30b1e493b1e4.4Standards ContributorsBroad community of developers, security researchers, and AI platform engineers providing feedback on the draft specification.Public ReviewerFiles issues, submits pull requests, and participates in Community Group calls to shape the specification before it advances.PerformerBeneficiary4.4.aPublic comments received and formally addressed before Draft advancesNumber of GitHub issues with documented resolution before advancing from Draftpi-4-4-aPLACEHOLDER_252027-06-30StatusResolved25At least 25 substantive GitHub issues resolved (closed with rationale, merged, or formally deferred) before the specification advances from Draft Community Group Report to Working Draft status.[To be determined]4.4.bDiversity of reviewer organizations providing substantive feedbackNumber of distinct organizations providing at least one resolved substantive commentpi-4-4-bPLACEHOLDER_262027-06-30StatusDiverse5At least 5 distinct organizations (including AI platform providers, browser vendors, web developer tool companies, and security research organizations) each provide at least one resolved substantive comment before the Draft advances.[To be determined]MCP InteroperabilityEnsure that WebMCP is architecturally complementary to — not competing with — Anthropic's Model Context Protocol (MCP), so that AI agents and AI platforms can interact with both server-side MCP tools and browser-side WebMCP tools through a coherent, interoperable ecosystem._c1e9f2e7-e8ff-4ff4-85f8-2d6a94772ba35AnthropicCreator of the foundational MCP standard on which WebMCP concepts are based; provider of Claude agents that are primary consumers of both MCP and WebMCP tools.Protocol Originator and AI Agent ProviderMaintains the MCP specification that WebMCP is designed to complement; ships Claude agents that benefit from coherent MCP/WebMCP interoperability.PerformerBeneficiaryOpenAIAI platform provider whose agents (ChatGPT, Agents SDK) support MCP and are expected consumers of WebMCP tools.AI Agent ProviderIntegrates WebMCP tool discovery into ChatGPT and Agents SDK; benefits from protocol alignment reducing per-platform integration work.BeneficiaryMCP Client DevelopersDevelopers building MCP-compatible agents and tools (Claude Desktop, OpenAI Agents SDK) who need WebMCP tools accessible through existing MCP infrastructure.Client IntegratorIntegrates MCP-B or native WebMCP transport into existing MCP client implementations to enable browser-side tool invocation.PerformerBeneficiaryRelationship: MCP (backend, server-side, JSON-RPC wire protocol) is the foundation. WebMCP adapts MCP concepts for the browser frontend without requiring a backend server. MCP-B bridges the two by translating WebMCP's browser-native format to MCP wire protocol, enabling existing MCP clients to call WebMCP tools. The three are complementary layers, not competing specifications.Protocol AlignmentMaintain conceptual and structural alignment between WebMCP tool definitions (name, description, inputSchema, execute) and MCP tool definitions so that developers familiar with MCP can adopt WebMCP with minimal additional learning, and so that tooling can serve both ecosystems._50b727ac-60a7-4a93-ba5e-2e77ed2f20da5.1W3C Web Machine Learning Community GroupResponsible for maintaining alignment between WebMCP tool definition fields and MCP tool definition fields during spec evolution.Alignment MaintainerReviews proposed WebMCP schema changes against MCP tool definition schema; documents divergences with rationale; coordinates with Anthropic on cross-protocol consistency.Performer5.1.aWebMCP tool definition fields mapped to MCP equivalents in specificationNumber of WebMCP tool fields with documented MCP mapping or documented divergence rationalepi-5-1-aPLACEHOLDER_272027-06-30StatusMapped5All five WebMCP tool definition fields (name, description, inputSchema, execute/handler, annotations) explicitly mapped to their MCP equivalents or accompanied by a normative rationale for any divergence, published in the specification.[To be determined]5.1.bDeveloper migration effort from MCP to WebMCP tool authoringMigration guide available (yes / no); estimated learning time (hours)pi-5-1-bPLACEHOLDER_2812027-12-31StatusPublished migration guide with estimated 2-hour learning curve for MCP-familiar developersA migration guide is published showing MCP-familiar developers how to author WebMCP tools, with an estimated onboarding time of 2 hours or less for developers already proficient in MCP server development. ^^ 1 = Available | 0 = Unavailable[To be determined]MCP-B CompatibilityEnsure that the MCP-B polyfill faithfully translates WebMCP tool registrations to MCP wire protocol (JSON-RPC), enabling existing MCP clients — Claude Desktop, OpenAI Agents SDK, and others — to discover and call browser-side WebMCP tools without modification to the client software._b90c3f84-0f51-47f2-a23a-4d575412242f5.2WebMCP-org CommunityMaintainers of MCP-B responsible for faithful protocol translation between WebMCP browser format and MCP JSON-RPC wire protocol.Bridge ImplementerImplements and maintains the translation layer between WebMCP's browser-native tool registration format and MCP's JSON-RPC wire protocol; validates against both specs.PerformerMCP Client DevelopersBenefit from MCP-B compatibility by calling WebMCP tools through their existing unmodified MCP client infrastructure.Unmodified Client BeneficiaryDiscovers and calls browser-side WebMCP tools through existing MCP clients (Claude Desktop, Agents SDK) without any client-side code changes.Beneficiary5.2.aMCP client operations successfully bridged to WebMCP tools via MCP-BPercentage of MCP tool operation types (list, call, describe) passing round-trip compatibility testspi-5-2-aPLACEHOLDER_292027-12-31StatusCompatible100100% of MCP tool operation types — list_tools, call_tool, and tool describe — pass round-trip compatibility tests between an unmodified MCP client and a WebMCP tool registered via MCP-B, with results verified against at least one conformant MCP client (e.g., Claude Desktop).[To be determined]5.2.bMCP transport types supported by MCP-BNumber of MCP transport types supported (tab transport and extension transport)pi-5-2-bPLACEHOLDER_302027-06-30StatusDocumented2MCP-B implements and documents both transport types: (1) tab transport (within-tab communication for browser-native agents) and (2) extension transport (cross-process communication via browser extension for external MCP clients).[To be determined]uuid-admin-webmcp-part22026-02-012028-12-312026-02-17https://starborn.github.io/webmcp/webmcp-complete-guide.htmluuid-submitter-owen-amburOwenAmburOwen.Ambur@verizon.net