Build a2a agents with Cursor.
a2a init writes an AGENTS.md at your repo root — exactly what Cursor's Agent reads for project instructions. Cursor opens already knowing the SDK: @a2a.tool, ctx.llm, durable files, the deploy flow. No rules to hand-write, no invented APIs, just working code against the real platform.
$ a2a init support-agent+ support-agent/agent.py+ support-agent/a2a.yaml+ support-agent/AGENTS.md# Cursor reads AGENTS.md at the repo root› Cursor Agent knows @a2a.tool, ctx.llm, workspace files› ⌘K a new skill against the real SDK, then a2a deploy
From init to deployed, with Cursor in the loop
Run a2a init
Scaffold the agent. The template writes an AGENTS.md at the repo root — the file Cursor reads — alongside your starter agent.py and manifest.
Build in Cursor
Cursor's Agent loads AGENTS.md and knows the SDK. ⌘K a new @a2a.tool skill, wire ctx.llm and the workspace, all against the real API.
Run a2a deploy
Verify with a2a validate and a2a card, then ship. You get a managed Postgres, an MCP server, an API, and a signed receipt per run.
Why teams build a2a agents in Cursor
The reference ships at the repo root where Cursor already reads, so its Agent is productive on the first prompt.
No rules to author
You don't hand-write .cursor rules to teach the SDK. a2a init drops an AGENTS.md where Cursor already looks.
One reference, every coding agent
AGENTS.md is the source of truth; CLAUDE.md and the .claude skill defer to it. Cursor, Codex, and Claude Code stay in sync.
Safe-by-default skills
The golden rules are in the reference — use ctx.llm, never read provider keys, durable files through the workspace backend — so Cursor's edits follow policy.
Verified before it ships
a2a validate and a2a card confirm the agent loads and its card is well-formed before the platform build runs.
“Cursor is already reading AGENTS.md. a2a just makes sure it says something useful.”
Before you deploy.
How does Cursor know how to build an a2a agent?
Cursor reads an AGENTS.md file at the repository root as project instructions for its Agent and inline edits. `a2a init` writes that file — a complete SDK reference — so Cursor opens your project already knowing the @a2a.tool decorator, the RunContext API, ctx.llm, durable file paths, and the deploy flow. No manual rules to author.
Do I need to set up .cursor/rules by hand?
No. The scaffolded AGENTS.md at the repo root is what Cursor reads, so you get an SDK-aware project out of the box without hand-writing rule files. If you later want Cursor-specific rules, you can add them, but the reference that makes Cursor productive ships with a2a init.
Will Cursor write real a2a code or hallucinate APIs?
Real code. Coding agents invent APIs when they have no source of truth; the shipped AGENTS.md is that source — actual decorator options, the ctx.llm credential rule (never read provider keys), the workspace-backed file paths. Cursor writes against the real SDK, then you verify with `a2a validate` and `a2a card` before deploying.
Does the same reference work in other tools?
Yes. AGENTS.md is a cross-tool standard: Cursor and Codex read it directly, and Claude Code gets a CLAUDE.md that imports it plus an invokable build skill. Whatever your teammates use, they all build against one consistent reference.
Let Cursor build your agent, then deploy it.
a2a init scaffolds a Cursor-ready agent in one command — AGENTS.md included — and a2a deploy ships it with a managed Postgres database, an MCP server, an API, a frontend, and an Ed25519-signed receipt for every run. The whole agent app, built in the editor you already use.