Suncly Verification · the mechanism behind every result

Your agent makes promises. See which ones it keeps.

Suncly tests A2A agents against their Agent Card and MCP servers against their tool list. You get a signed report of what passed, what failed and what was never tested.

Start a test See a sample report

Priced from your tool list before you start · card held, charged when the report is delivered

Want a monthly answer for your users’ tasks? See Benchmark →
Judged on what happened, not on what the agent said
01
Every test repeated
02
Writes tested too, with your approval
03
Up to 7 protocol checks on each run
04
Checked against 11 lab agents and the A2A reference agents
05

Where it fits

One workflow, three moments.

01

Before you ship

Verify your agent against its own Agent Card, or your MCP server against its tool list, before a release or a directory submission.

02

Before you connect

Verify someone else’s agent or server, with the owner’s permission, before it gets access to your systems.

03 · early access

Every month after

Benchmark runs your users’ tasks in ChatGPT, Claude and Copilot every month, and shows what changed. See Benchmark

Who it’s for

I run an MCP server Tested against every tool you list, the way an app-directory reviewer calls them.
I build an A2A agent Tested against every skill your Agent Card declares.
I approve agents for my company Evidence for a vendor review, before the agent gets access.
I own an agent in production Your users’ tasks, run in every client each month, with the cause of every failure. That’s Benchmark.

What you get

Kept Hold Broken Not tested
Harbor Returns Agent
v2.3.1 · A2A 1.0 · 20 runs · sandbox
Altitude 47 · Hold · 3 of 4 skills tested
order-status
Tells a customer where an order is and when it will arrive.
8 of 10
start-return
Opens a return and emails a prepaid label.
0 of 5
refund-estimate
Estimates the refund, including restocking fees.
5 inconcl.
cancel-order
Declares no examples.
not tested
Signed · ed25519-0d7631d77bb7b9a6Card hash, plan, counts and transcripts

An excerpt of a real Suncly report for a fictional agent: every claim, every run and its verdict, and what was never tested.

Open the full sample report →
Writes are tested, not skippedCreate tools run under the plan you approve, on a test account. Update and delete only ever touch objects we created and can prove are ours. In a recent MCP server report, three create tools ran and kept their promise, and every tool we didn’t call, including update, delete and send, was listed by name. A run that stays read-only says so on its first page.
Paper squares in green, mustard and terracotta laid out in a grid on linen

Why now

150+organisations support A2A, and it ships inside the major cloud agent platforms.Linux Foundation, Apr 2026 ↗
40%of working MCP servers in public registries expose a destructive or execute tool.PolicyLayer, May 2026 ↗
91.8%of 414 internet-facing MCP servers tested had no OAuth.arXiv:2608.00150 ↗

A signed Agent Card proves who published it. Suncly shows whether the agent does what it says.

Pricing

Priced from your server, shown before anything runs.

See pricing
01

Verification

One server or agent, one signed report. Paste a URL and see the price in the console.

Priced from your tool list: read tools cost least, write and destructive tools more, because they take more care to test safely. You see your exact price before anything runs, and you’re charged only when your report is delivered.

Get a price
02

Team

For teams that ship every week. A report on each release, first in the queue.

Talk to us
03

Enterprise

For platform and security teams, AI labs and registries. Custom plans, DPA, invoice or PO.

Talk to us
Benchmark · early access

Next: does it work for your users, every month?

Verification shows whether a server or agent keeps its own promises. Benchmark is the answer an owner needs on top of that, measured every month.

See Benchmark
BenchmarkYour users’ tasks run end to end in ChatGPT, Claude and Copilot, twenty times each, every month. A pass rate with its N, every failure tagged by cause, and the change since last month.
Verification add-ons · in development
Card WatchA daily check of your Agent Card and tool list. A change drafts a new plan for a person to approve.
Release gate in CIVerify every release in your pipeline before it ships.
Registry and badgeA public record of each verification, and a badge that links to it.
The sun rising over the sea

Bring one agent. Leave with evidence.

Price shown before you start. Charged when your report is delivered.