Awesome Agent Protocols

A curated list of protocols and standards for AI agents.

🔌 wire protocol · 📐 spec / standard · 📄 format / convention · ⚠️ deprecated or merged

Contents

• Foundations
• Tool and context
• Agent to agent
• Agent to UI
• Agent to editor
• Serving and runtime
• Identity and discovery
• Agentic web
• Payments and commerce
• Formats and conventions
• Governance
• Surveys
• Examples

Foundations

The editor-tooling protocols that predate agents. Same idea — decouple client from provider over JSON-RPC, collapse M×N to M+N.

• Language Server Protocol (LSP) 🔌 - JSON-RPC standard for editor↔language-intelligence communication. The original M+N protocol; everything below copies this pattern. protocol
• Debug Adapter Protocol (DAP) 🔌 - JSON-RPC for editor↔debugger, through an intermediary adapter. protocol
• Build Server Protocol (BSP) 🔌 - Scala Center / JetBrains JSON-RPC for IDE↔build-tool (targets, compile, test, run). protocol

Tool and context

How an agent calls tools, reads data, and gets context.

• Model Context Protocol (MCP) 🔌 - Anthropic. JSON-RPC 2.0 over stdio or Streamable HTTP. Connects LLM hosts to tools, resources, and prompts. Now under the Linux Foundation. protocol
  • Tools 🔌 - tools/list + tools/call. JSON-Schema-typed callable functions. protocol
  • Resources 🔌 - URI-addressed readable context (files, rows, blobs) with subscriptions. protocol
  • Prompts 🔌 - Server-defined parameterized prompt templates, surfaced as slash commands. protocol
  • Sampling 🔌 - Server asks the client for an LLM completion. Supports tool use + toolChoice. protocol
  • Elicitation 🔌 - Server pauses to ask the user for input (forms) or hand off to the browser (OAuth, payments). protocol
  • Roots 🔌 - Client declares which file:// paths servers may touch. protocol
  • Apps (SEP-1865) 🔌 - Servers ship interactive HTML UI as ui:// resources, rendered in a sandboxed iframe. protocol
  • Registry 📐 - Public server catalog + REST API. Standardized server.json publish format. spec/standard
• OpenAI Apps SDK 🔌 - ChatGPT app = server logic + embedded widget. Converging with the MCP Apps extension. protocol
• agents.json 📐 - Wildcard AI. Layered on OpenAPI — adds flows, links, and auth so agents can drive REST APIs. spec/standard
• NLWeb 🔌 - Microsoft. Turns a website into a conversational, agent-queryable endpoint. protocol
• Eclipse LMOS 🔌 - Deutsche Telekom / Eclipse Foundation. W3C Web of Things-based agent + tool discovery. protocol
• SLOP 🔌 - Six REST endpoints (chat, tools, memory, resources, pay, info). Deliberately minimal. protocol
• Arazzo 📐 - OpenAPI Initiative. YAML/JSON format for deterministic multi-step API workflows. spec/standard

Agent to agent

Cross-vendor, cross-framework agent collaboration.

• Agent2Agent (A2A) 🔌 - Google / Linux Foundation. JSON-RPC 2.0 + SSE + Agent Card discovery. Stateful async task delegation between peer agents. protocol
• Agent Communication Protocol (IBM ACP) 🔌 ⚠️ - IBM / BeeAI. REST-native agent messaging. Merged into A2A (Aug 2025). protocol
• Agent Network Protocol (ANP) 🔌 - Decentralized P2P. W3C did:wba authentication, JSON-LD descriptions. protocol
• Agora 🔌 - Oxford. Agents speak natural language for rare messages, then negotiate self-generated protocol docs for frequent ones. No central registry. protocol
• AITP 🔌 - NEAR AI. Cross-trust-boundary Threads with pluggable capabilities for payments, decisions, on-chain identity. protocol
• Coral 🔌 - MCP-native coordination. @mention-addressed threads, on-chain micropayments. protocol
• HCS-10 OpenConvAI 🔌 - Agent messaging over Hedera Consensus Service topics. Tamper-proof, broker-less. protocol
• ACNBP 🔌 - Draft academic. Discovery → negotiation → binding handshake with PKI signatures. protocol
• W3C AI Agent Protocol CG 📐 - W3C community group for decentralized agent identity and discovery. Ships ANP. spec/standard

Agent to UI

Backend → frontend streaming of agent state, tool calls, and thinking.

• AG-UI 🔌 - CopilotKit. Typed JSON events over HTTP/SSE from any agentic backend to any frontend. 9k+ GitHub stars. protocol
• A2UI 🔌 - Google. Declarative generative-UI: JSON-Lines surface + data-model updates against a client widget catalog. protocol
• AI SDK UI Stream Protocol 🔌 - Vercel. SSE of typed message parts (text, reasoning, tool, data) consumed by useChat. protocol

Agent to editor

The "LSP for agents" — editor/IDE ↔ coding agent.

• Agent Client Protocol (ACP) 🔌 - Zed. JSON-RPC for editor↔agent. Adopted by Zed, JetBrains, VS Code, Kiro, Cline. Spoken by Gemini CLI, Claude, Codex. protocol
• LSAP 🔌 - Turns LSP capabilities into high-level agent-native "cognitive" tools. protocol
• Cody Agent Protocol 🔌 ⚠️ - Sourcegraph. Bidirectional JSON-RPC over stdio. Effectively frozen. protocol

Serving and runtime

How an agent is invoked as a service — runs, threads, state.

• LangChain Agent Protocol 📐 - Framework-agnostic REST/OpenAPI. Agents, Runs, Threads, Store endpoints. spec/standard
• Agent Connect Protocol (ACP) 📐 - AGNTCY / Cisco. OpenAPI spec for remote agent invocation, any framework. spec/standard

Identity and discovery

How agents are named, found, and trusted.

• Agent Name Service (ANS) 📐 - OWASP. DNS-inspired discovery with PKI identity. Protocol-agnostic adapter over A2A, MCP, ACP. spec/standard
• Project NANDA 📐 - MIT Media Lab. Federated "DNS for agents" — human-friendly handles → signed pointers → verifiable AgentFacts. spec/standard
• AGNTCY Directory 🔌 - Cisco. Distributed, content-addressed registry. Signed OASF records as OCI artifacts over libp2p DHT. protocol
• OASF 📐 - AGNTCY. OCI-based schema describing agent attributes, skills, capabilities. spec/standard
• ERC-8004 📐 - On-chain agent identity, reputation, and validation via three Ethereum registry contracts. spec/standard
• TWZRD Agent Intel 🔌 - Solana-native x402 MCP for agent trust scoring and identity verification. On-chain reputation scoring via HTTP 402 + USDC micropayments; signed twzrd.receipt.v5 trust tokens for verifiable agent identity. tool/service
• SPIFFE / SPIRE 📐 - CNCF. Short-lived cryptographic workload identities, no secrets at rest. Increasingly used for agent runtime identity. spec/standard
• AgentDNS 📐 ⚠️ - Expired IETF draft. DNS-inspired naming + semantic discovery. Small reference impl. spec/standard
• DNS-AID 📐 - IETF DNSOP draft. Publish/discover/verify agents through standard DNS records. Linux Foundation ref stack. spec/standard

Agentic web

Agents on the web — page tools, crawl permissions.

• WebMCP 🔌 - W3C (Google, Microsoft). navigator.modelContext browser API exposing in-page tools to agents. protocol
• Web-Agent Protocol (WAP) 📄 - Record browser interactions via extension, replay them for autonomous web agents. format/convention
• TDMRep 📐 - W3C. Machine-readable mining permissions in HTTP headers, .well-known, or HTML. spec/standard

Payments and commerce

How agents pay — authorization, checkout, stablecoins.

• AP2 🔌 - Google + 60 partners. A2A extension. Cryptographically signed Mandates for agent-led payments, cards + stablecoins. protocol
• x402 🔌 - Coinbase / Linux Foundation. HTTP 402 with stablecoin payment in a header. Pay-per-call APIs. protocol
• Agentic Commerce Protocol (ACP) 🔌 - OpenAI + Stripe. Agent-led checkout with Shared Payment Tokens. Powers Instant Checkout in ChatGPT. protocol
• Universal Commerce Protocol (UCP) 📐 - Google + Shopify (Etsy, Target, Walmart, Wayfair). Product discovery through checkout. spec/standard
• Cloudflare Pay Per Crawl 🔌 - HTTP 402 micropayments for AI crawlers, per request. protocol

Formats and conventions

Files, not messages. Often called "protocols" — they're not.

• AGENTS.md 📄 - Project-specific instructions for coding agents. 60k+ repos. Agentic AI Foundation (Linux Foundation). format/convention
• Agent Skills (SKILL.md) 📄 - Anthropic. Reusable agent capability as YAML frontmatter + instructions + optional scripts. format/convention
• llms.txt 📄 - Markdown at site root listing LLM-readable content. Fetched by coding agents, not yet honored by major crawlers. format/convention
• ai.txt 📄 - Usage permissions for AI (no-training, no-inference, allow-RAG). Sits next to robots.txt. format/convention
• JSONL transcripts 📄 - One JSON object per line. How Claude Code, Codex, and others persist sessions. A log format, not a wire protocol. format/convention

Governance

Who stewards the open agent stack.

• Agentic AI Foundation 📐 - Linux Foundation. Co-founded by Anthropic, Block, OpenAI. Homes MCP, AGENTS.md, Goose. spec/standard
• Linux Foundation — A2A 📐 - Hosts A2A (+ absorbed IBM ACP), AP2, and related infra. spec/standard
• OpenAPI Initiative 📐 - OpenAPI + Arazzo. The schema layer agent protocols build on. spec/standard

Surveys

• A Survey of AI Agent Protocols - Two-dimensional taxonomy (context-oriented vs inter-agent, general vs domain-specific). Covers ANP, Agora, AITP, Coral.
• A Survey of Agent Interoperability Protocols - Side-by-side of MCP, ACP, A2A, ANP across nine dimensions, with a phased adoption roadmap.

Examples

Wire-message examples in examples/: MCP tool calls, A2A Agent Cards, agents.json flows, LangChain Runs/Threads, x402 payments, AP2 mandates.

Related

• Awesome MCP Servers
• Awesome A2A
• Awesome Agentic Payments

Skills

This repo ships a Claude Code skill that uses the catalog during development — recommends protocols, pulls wire examples, checks compliance, surfaces contribution opportunities.

Contributing

Read contributing.md first. By participating you agree to the Code of Conduct.