AGENT2AGENT
SuperQode A2A Agent
An agent other agents can ask which coding agent to use.
It publishes an Agent Card, answers Agent2Agent requests over JSON-RPC and HTTP+JSON, and returns a ranked shortlist of third party coding agents and harnesses from the Harness Hub. SuperQode’s own harnesses are excluded from the ranking and disclosed separately, so the answer stays vendor neutral. No repository is sent.
TCK: 100% MUST and 100% MAY on the JSON-RPC and HTTP+JSON transports the card declares, 96.4% overall.
LIVE SHORTLIST
Ask the hosted agent
One sentence. No repository is sent. Open-tier ranking from the Hub.
THREE WAYS TO RUN IT
Yours, ours, or yours in your cloud
The software is Apache 2.0 in every case.
What is paid for is operation, the model calls behind a keyed request, and a catalogue kept current. If you would rather run it yourself, the same server ships in the open source package and always will.
Self-host
Run it yourself, with your own secret and your own model.
- +Ships in the open source package: superqode serve a2a
- +Mint your own keys with superqode a2a-keys issue
- +Your model provider, your key, your bill
- +Answers from the public Harness Hub catalogue
- +Apache 2.0, no limits we impose
Hosted
We run it at a2a.superqode.dev, so you call an endpoint.
- +Open tier is live now: catalogue shortlist, keyword ranked, no key needed
- +Keyed access adds model-read requests and a higher rate limit
- +Keys are signed, carry a customer, a tier and an expiry
- +Catalogue kept current, so callers do not track upstream changes
- +Registrable with Gemini Enterprise, Microsoft Foundry and Bedrock AgentCore
Private
The same agent, deployed inside your own cloud.
- +Runs in your network, so nothing leaves it
- +Your model provider and your own data boundary
- +Private Hub entries for internal harnesses, alongside the public catalogue
- +Your own rate limits, key policy and audit
Keyed access is not switched on yet. The open tier at a2a.superqode.dev is live and answering today. Signed keys work in the software, and self-hosting supports them now, but the hosted deployment has no signing secret configured, so a keyed call is refused rather than served. Ask for a key and we will tell you when it opens.
WHAT A KEY CHANGES
With a key, and without
The difference is how your request is read.
| Open | Keyed | |
|---|---|---|
| Request read by | Keyword parser | A model |
| Rate limit | 10 a minute | 60 a minute |
| Credential | None | sqk_live_... |
| Catalogue | Full Hub | Full Hub |
| Repository sent | Never | Never |
A caller passes a key as SUPERQODE_A2A_TOKEN. Keys are signed rather than stored, so verifying one is a signature check and a clock comparison.
CALL IT
$ superqode connect a2a --url https://a2a.superqode.dev$ superqode connect a2a --conformance --no-send$ superqode serve a2a --spec harness.yaml$ superqode a2a-keys issue "Platform Team" --tier standard --days 30DISCOVERY
The Agent Card declares JSON-RPC and HTTP+JSON across A2A 1.0 and 0.3 from one endpoint, so a single document satisfies every registration path.
Ask for a key
Tell us what you want to call it from and we will tell you when keyed access opens. If you would rather not wait, the same server is in the repository and self-hosting works today.