--- title: "Standards and Protocols" description: "Portable standards, discovery formats, and MCP tooling for agent surfaces" url: "https://agentsurface.dev/docs/tooling-catalog/standards-and-protocols" lastVerified: 2026-09-24 lastModified: 2026-09-25T05:46:01.000Z --- Use these when agent surfaces need to be discoverable, inspectable, portable, or authenticated without relying on vendor-specific prompt conventions. ## Standards and Protocols [#standards-and-protocols] Prioritize these because they outlive individual vendors and keep agent surfaces portable. | Tool | Website | What it is for | Tier | Why it matters | | ------------------------------------------------------------------------------------------------------------- | -------------------------------- | ----------------------------------------------------------------------------------- | ----------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | [Model Context Protocol](https://modelcontextprotocol.io/docs/getting-started/intro) | modelcontextprotocol.io | Standard protocol for exposing tools, resources, prompts, and apps to AI clients | Core | The strongest current interoperability bet for agent tools and external context. | | [MCP Registry](https://registry.modelcontextprotocol.io) | registry.modelcontextprotocol.io | Discoverable registry for MCP servers | Conditional | Useful for distribution, but still evaluate server quality case by case. | | [MCP Server Cards](https://github.com/modelcontextprotocol/modelcontextprotocol/pull/427) | github.com/modelcontextprotocol | Proposed `.well-known` metadata for discovering MCP servers before connecting | Watch | Worth tracking because it makes tool endpoints discoverable without scraping docs. | | [AGENTS.md](https://agents.md) | agents.md | Repository instructions for coding agents | Core | Simple, durable context format for Codex, Claude Code, Cursor, Copilot, Devin, and similar tools. | | [AgentReady](https://agentready.org) | agentready.org | MIT-licensed, vendor-neutral spec of stable-ID requirements for agent find/read/act | Watch | v1.0 (August 2026), developed by Ora, Vercel, Mintlify, and Auth0. Defines MUST/SHOULD/MAY requirements (`AR-FIND-*`, `AR-READ-*`, `AR-ACT-*`) but deliberately no scoring; see the [scanner comparison](/docs/scoring/public-scanners#ora--orank-scans). | | [Agent Skills](https://agentskills.io/home) | agentskills.io | Reusable bundles of instructions, scripts, references, and assets for agents | Core | Good abstraction for repeatable specialized capability without bloating the base prompt. Use the public spec for `SKILL.md` shape and validation. | | [llms.txt](https://llmstxt.org) | llmstxt.org | LLM-readable site and documentation index | Conditional | Useful when publishing docs for agents to retrieve without noisy crawling. | | [Markdown content negotiation](https://developers.cloudflare.com/fundamentals/reference/markdown-for-agents/) | developers.cloudflare.com | Serving clean Markdown when clients request `Accept: text/markdown` | Core | Lowers token waste and gives agents a readable version of public pages. | | [Content Signals](https://contentsignals.org) | contentsignals.org | Robots-compatible declarations for search, AI input, and AI training permissions | Conditional | Gives agent crawlers more specific content-use policy than allow/block alone. | | [Web Bot Auth](https://datatracker.ietf.org/doc/draft-meunier-web-bot-auth-architecture/) | datatracker.ietf.org | Draft standard for bots to identify themselves with signed HTTP requests | Watch | Important if sites need to distinguish friendly agents from generic automation. | | [API Catalog](https://www.rfc-editor.org/rfc/rfc9727.html) | rfc-editor.org | `.well-known` catalog for public API discovery | Watch | Useful for services with multiple APIs, specs, docs, and status endpoints. | | [JSON Schema](https://json-schema.org) | json-schema.org | Schema language for structured inputs and outputs | Core | Foundation for tool schemas, structured output, eval fixtures, and API validation. | | [OpenAPI 3.1](https://spec.openapis.org/oas/v3.1.0) | spec.openapis.org | Machine-readable HTTP API contracts | Conditional | Include when turning existing APIs into safe, typed agent tools. | | [Arazzo](https://spec.openapis.org/arazzo/latest.html) | spec.openapis.org | Multi-step API workflow descriptions | Watch | Promising for documenting tool sequences that agents should not infer from endpoints alone. | | [OAuth protected resource metadata](https://www.rfc-editor.org/rfc/rfc9728.html) | rfc-editor.org | Discovery metadata for OAuth-protected resources | Conditional | Helps agents find the right authorization server instead of borrowing browser sessions. | | [OpenTelemetry GenAI](https://opentelemetry.io/docs/specs/semconv/gen-ai/) | opentelemetry.io | Semantic conventions for GenAI spans and events | Core | Gives agent traces a common shape across model calls, tools, retrieval, and evaluations. | | [AG-UI](https://docs.ag-ui.com) | docs.ag-ui.com | Agent-user interaction protocol for frontend and backend agent state | Watch | Good signal for standardizing event streams between agent backends and user-facing apps. | | [x402](https://www.x402.org) | x402.org | HTTP-native payments for machine and agent access | Watch | Relevant when agents need to pay for data, tools, APIs, or content without human checkout. Productized by Cloudflare Monetization Gateway (waitlist, July 2026). | | [Universal Commerce Protocol](https://ucp.dev) | ucp.dev | Agentic commerce discovery and transaction protocol | Watch | Track for agent-driven shopping and commerce surfaces. | | [Agentic Commerce Protocol](https://www.agenticcommerce.dev) | agenticcommerce.dev | Commerce protocol for agent-mediated purchasing | Watch | Beta but contracting: OpenAI shut down Instant Checkout in March 2026 and momentum has moved to UCP. Relevant only to maintain an existing ACP integration. | [DNS-AID](https://datatracker.ietf.org/doc/draft-mozleywilliams-dnsop-dnsaid/) is an emerging discovery proposal, currently an individual Internet-Draft. Consider it only for compatible agent clients; see [DNS Discovery](/docs/discovery/dns-discovery). ## Agent-Readable Web and Access Control [#agent-readable-web-and-access-control] Use these when the site itself should be easy for agents to discover, read, authenticate against, or audit. | Tool | Website | What it is for | Tier | Notes | | -------------------------------------------------------------------------------------------------------------------------- | ------------------------- | ---------------------------------------------------------------------- | ----------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | [Agent Readiness](https://isitagentready.com/) | isitagentready.com | Lighthouse-style scanner for how well a public site supports AI agents | Core | Directly relevant to this project. Includes DNS-AID and separate Content Signals checks. See the [scanner comparison](/docs/scoring/public-scanners#cloudflare-agent-readiness) for coverage and applicability. | | [Cloudflare URL Scanner Agent Readiness](https://blog.cloudflare.com/agent-readiness/) | blog.cloudflare.com | Programmatic site scanning with an agent-readiness report | Conditional | Useful when readiness checks should run in audits, CI, or recurring monitoring. | | [Markdown for Agents](https://developers.cloudflare.com/fundamentals/reference/markdown-for-agents/) | developers.cloudflare.com | Managed conversion of public pages into Markdown for agents | Conditional | Good reference pattern for making documentation cheaper and more reliable for agents to read. | | [AI Crawl Control](https://developers.cloudflare.com/ai-crawl-control/) | developers.cloudflare.com | Visibility, controls, and policy management for AI crawlers | Conditional | Relevant for publishers, docs sites, and products that need to know which AI services access content. | | [Managed robots.txt for AI crawlers](https://developers.cloudflare.com/bots/additional-configurations/managed-robots-txt/) | developers.cloudflare.com | Managed `robots.txt` directives and Content Signals for AI bot traffic | Conditional | Good operational path when site owners want crawl policy without hand-maintaining every directive. | | [Cloudflare Radar AI Insights](https://radar.cloudflare.com/ai-insights) | radar.cloudflare.com | Internet-wide data on AI crawler and agent-standard adoption | Watch | Useful for tracking which agent-readable web standards are gaining real adoption. | ## MCP Tooling [#mcp-tooling] Use these when building agent-accessible tools rather than one-off function calls. | Tool | Website | What it is for | Tier | Notes | | ---------------------------------------------------------------------------- | ------------------------------- | --------------------------------------------------------------------------------------------------- | ----------- | ------------------------------------------------------------------------------------------------------------------------------------------- | | [MCP TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk) | github.com/modelcontextprotocol | Build MCP clients and servers in TypeScript | Core | Primary SDK for TypeScript MCP work. | | [MCP Python SDK](https://github.com/modelcontextprotocol/python-sdk) | github.com/modelcontextprotocol | Build MCP clients and servers in Python | Conditional | Use when Python owns the tool boundary. | | [MCP Inspector](https://modelcontextprotocol.io/docs/tools/inspector) | modelcontextprotocol.io | Local inspection and debugging for MCP servers | Core | Essential for validating tool descriptions, schemas, resources, and prompts. | | [mcp-handler](https://github.com/vercel/mcp-handler) | github.com/vercel | Host MCP servers in web runtimes such as Next.js | Core | Good fit for TypeScript web apps that need to expose MCP endpoints. | | [FastMCP](https://github.com/jlowin/fastmcp) | github.com/jlowin | Python-first MCP framework | Conditional | Useful when decorators and Python service boundaries are simpler than raw protocol plumbing. | | [OpenAI Apps SDK](https://developers.openai.com/apps-sdk/) | developers.openai.com | Build MCP-backed apps and UI inside ChatGPT | Watch | Important direction for ChatGPT-integrated tools; track platform maturity and review requirements. | | [Smithery](https://smithery.ai/docs) | smithery.ai | MCP registry and hosted connection layer | Watch | Useful for discovery and connection workflows, but inspect each server's trust boundary, auth model, and maintenance state before adoption. | | [Klavis AI](https://docs.klavis.ai/) | docs.klavis.ai | Hosted and open-source MCP integration layer with managed authentication and tool-format conversion | Conditional | Good when an agent product needs many SaaS tools through MCP without building every OAuth and connector path itself. | | [TurboMCP](https://turbomcp.ai/) | turbomcp.ai | Self-hosted MCP gateway and management platform with server registry, RBAC, DLP, and audit logging | Watch | mcp.run now redirects here. Evaluate self-hosting requirements before adoption. | | [Ampersand MCP Server](https://docs.withampersand.com/mcp) | docs.withampersand.com | Hosted MCP server over Ampersand's connector platform | Watch | Alpha according to its own docs; evaluate for embedded integration products rather than as a default MCP layer. | Sources are linked inline.