Open protocols and proposed formats for how agents connect to tools, code editors, websites, payments, checkout and each other.
7 entries. Latest source check .
What a source check covers
Documentation checked means the source was read. It does not mean the software was installed, tested or security audited.
Reading an entry does not authorise installation, credential access or spending. Tech Horizon Labs projects are labelled, and there are no affiliate links or paid placement.
People building hosts, clients or servers that speak MCP.
Access and use
Spec and docs at modelcontextprotocol.io. Official org is github.com/modelcontextprotocol. Python SDK (MIT): uv add "mcp[cli]" or pip install "mcp[cli]", Python 3.11 or later. TypeScript SDK at github.com/modelcontextprotocol/typescript-sdk. Public servers are listed in the Official MCP Registry already in this catalog. The servers repo keeps only steering-group reference servers. New spec and code contributions are Apache-2.0; some older MIT remains; docs excluding the spec are CC BY 4.0.
People connecting agents built on different frameworks.
Access and use
pip install a2a-sdk. Also official Go, JavaScript, Java, .NET and Rust SDKs. Spec v1.0 at a2a-protocol.org. Linux Foundation project, donated by Google. Apache-2.0.
People wiring an editor such as Zed or JetBrains to a coding agent, or shipping either side of that pair.
Access and use
Spec at agentclientprotocol.com. Current stable protocol version is 1. TypeScript: npm install @agentclientprotocol/sdk. Rust: cargo add agent-client-protocol. Also official Python, Java and Kotlin SDKs. Apache-2.0.
A proposed Markdown file at /llms.txt, or at any path on a site, that gives agents a short summary of the site and links to clean Markdown versions of its pages.
Site owners and documentation teams who want coding agents and assistants to find the right pages, and people writing tools that read those files.
Access and use
A plain file you publish. It grants no access, and agents read it on demand while they help a user. Format: an H1 name, an optional blockquote summary, then H2 sections of Markdown link lists, with an Optional section for links an agent can skip. Version 2 also proposes .md copies of pages, linked with rel="alternate" type="text/markdown" and rel="describedby". Written by Jeremy Howard and open for community input at github.com/AnswerDotAI/llms-txt, which is Apache-2.0.
An open standard for AI agents, apps, businesses and payment providers to handle product search, carts, checkout, identity linking and orders without a custom integration for each connection.
Retailers, commerce platforms, payment providers and AI platforms that want agents to buy from businesses through one interface.
Access and use
Spec and docs at ucp.dev; schemas, SDKs, samples and conformance tests at github.com/Universal-Commerce-Protocol. Businesses declare the capabilities they support in a profile and offer them over REST, MCP or A2A. Identity linking uses OAuth 2.0, and payments can use AP2 mandates. Businesses stay the merchant of record. Co-developed by Google, Shopify, Etsy, Wayfair, Target, Walmart and others. The latest dated release is v2026-08-25; the lodging spec is a draft. An agent that completes a checkout can spend money, so set limits. Apache-2.0.
An open standard from OpenAI and Stripe for checkout between buyers, their AI agents and businesses, including passing payment credentials through the agent without exposing them.
Businesses that want their checkout to work with ChatGPT and other agents, AI platforms that add purchasing, and payment providers.
Access and use
The spec is in beta, with dated versions; the latest stable one is 2026-04-17, published as RFCs, OpenAPI and JSON Schema files on GitHub. It works over REST or MCP. ChatGPT was the first platform to implement it, and businesses apply to sell there. Stripe was the first compatible payment provider, through its Shared Payment Token. Businesses stay the merchant of record and can accept or decline each agent transaction. An agent that completes a checkout can spend money, so set limits. Apache-2.0.