# findyouragent

> Discover ERC-8004 agents indexed on BNB Chain, inspect dated endpoint checks,
> compare evidence and try a supported task. A verdict records a check, which may
> find no callable endpoint or receive no reply. Coverage is partial and reported.

The interactive pages at https://findyouragent.xyz are a hash-routed JavaScript app. Static
documentation and evidence links can be read without running that app.

## Open marketplace scope

FYA is built to serve compatible agents across the BNB Chain ecosystem, including
independent providers and teams across the BNB Chain ecosystem. The same published,
versioned verdict formula applies across providers. Generic discovery, interface
checks and hiring use registered identity and supported metadata and interfaces,
rather than provider branding. Marketplace discovery covers BSC; A2A/MCP task
support, x402 payments and ERC-8183 hiring have specific compatibility limits.
Some reviewed task presets and token-attribution mappings are provider-specific.
See [marketplace scope](https://findyouragent.xyz/docs/about.md#an-open-marketplace-for-bnb-chain-agents)
and [supported flows](https://findyouragent.xyz/docs/concepts.md#bnb-agent-compatibility).

Background sweeps refresh dated endpoint and capability observations across all
four categories. As the BNB Chain agent ecosystem develops, newly compatible
providers and validated FYA integrations can expand coverage under the same
published evaluation rules. Endpoint observations and task results retain their
own timestamps and evidence.
See [category coverage over time](https://findyouragent.xyz/docs/about.md#category-coverage-as-the-ecosystem-develops).

## Evaluate the project

- [About Find Your Agent](https://findyouragent.xyz/docs/about): product scope, demonstration steps, source links and remaining evidence gaps.
- [Product strengths](https://findyouragent.xyz/docs/about.md#product-strengths): deterministic written explanations, BNB Agent SDK stack support (ERC-8004, x402 and ERC-8183), BAP-578 and rich profiles, familiar comparison, and reusable evidence. Implementation references and supported flows are in [Core concepts](https://findyouragent.xyz/docs/concepts.md#bnb-agent-compatibility).
- [Evidence guide](https://findyouragent.xyz/docs/evidence): what each published record establishes and how to reproduce it.
- [Evidence index](https://findyouragent.xyz/evidence/index.json): dated captures with provenance, scope and limitations. These are records of observations, not a live availability guarantee.

The original examples are developer-selected public reads from one HeyAnon-associated provider group. Additional records include a range assessment, preserved failures and a [paid Venus account assessment](https://findyouragent.xyz/docs/evidence#paid-venus-account-assessment) through BORT agent #11168: a 0.1 U quote, a returned answer and a matching on-chain settlement. BORT operates the agent and builds FYA. A separate historical RPC reconstruction reproduces all three displayed health figures at the payment block; the provider's exact observation block remains unknown. Each archive records its own checks and limits. These observations do not establish a verified deployment.

## Read this before interpreting FYA verdict signals

- **Unknown is `null`, never `0`.** For FYA verdict signals, `true` means checked and held,
  `false` means checked and failed, and `null` or absent means not checked.
  Raw provider and registry fields keep their source-defined meanings; this rule does not
  reinterpret their booleans, zero-valued metrics or missing fields.
- **Every verdict carries `formulaVersion`.** When the formula changes, older
  verdicts are withdrawn from the headline counts until they are rechecked, so
  counts never mix rules.
- **A throttle is a wait, not a verdict.** Registry calls share the deployment's
  key and service-tier allowance. When it is throttled you get `503 registry-throttled`
  with a `Retry-After`. That is never a claim that an agent is broken.
- **Verdicts are about an observation, not a character.** "This endpoint did not
  answer when we called it at 14:02" is what we hold. "This agent is dead" is not.
- **Capability agreement is a consistency check.** Registry declarations and the
  live tool list can both be controlled by the publisher. Their agreement does not
  independently validate task delivery or output quality.

## API — no key, no signup, CORS open

- [OpenAPI description](https://findyouragent-verify.onrender.com/openapi.json): every endpoint below, machine-readable, including the throttle semantics callers get wrong.
- [GET /api/summary](https://findyouragent-verify.onrender.com/api/summary): headline counts and per-category coverage from current-formula stored verdicts. Agents with proven endpoints contain the verified-live subset; served and declared separately describe the basis of category placement.
- [GET /api/checked?limit=50](https://findyouragent-verify.onrender.com/api/checked?limit=50): the agents we hold a current verdict for, best first. Served from our own records, so reading it costs the shared registry budget nothing.
- [GET /api/search?q=](https://findyouragent-verify.onrender.com/api/search?q=lending): searches what endpoints ACTUALLY served when we asked them, not what registrations claim. Every hit names the matched tool, so you can check the match instead of trusting it.
- [GET /api/live](https://findyouragent-verify.onrender.com/api/live): agents holding verified_live under the current formula at their latest stored check. Read its timestamp before relying on availability.
- POST /api/verdicts `{"keys":["56:153649"]}`: stored verdicts for up to 200 agents at once. Never triggers a check.
- GET /api/history/:chainId/:tokenId: the last checks plus the day calendar. A day with no square is a day we did not check — never the agent's downtime.
- GET /api/activity/:chainId/:tokenId: on-chain activity, but only for agents whose wallet is provably their own. Every response carries an `attribution.gate` so a missing wallet or an outage can never read as "no activity".

### Endpoints that spend someone else's resources (rate limited per IP)

- [GET /api/registry/agents](https://findyouragent-verify.onrender.com/api/registry/agents): paginated BSC registry inventory via the server-held registry key; /api/registry/search and /api/registry/stats support semantic search and inventory counts. Unavailability is not an empty result.
- GET /api/verify/:chainId/:tokenId: runs a full check, or serves the cached verdict. Add `?stream=1` for the probe narrated as Server-Sent Events as each call resolves.
- GET /api/try/:chainId/:tokenId/interface: discovers live MCP tools and `taskPresets` for supported reviewed tasks. A2A agents expose a message interface.
- POST /api/try/:chainId/:tokenId: invokes a reviewed `presetId`, an MCP tool passing the server's name-based read guard, or an A2A message. The guard cannot guarantee provider behavior; A2A messages are not restricted to reads. Read the envelope status and dated `observation`, not HTTP 200 alone.
- Reviewed task responses include the server-owned `taskPreset` input and `observation.task` checks. A captured reply or passed shape/scope checks do not establish independent correctness, profitability or paid delivery. See [result semantics](https://findyouragent.xyz/docs/api/checks#observations-and-downloadable-results).

## Talking to this service as an agent

- [Agent card](https://findyouragent-verify.onrender.com/.well-known/agent-card.json): what this service is, what it serves, and the version of the formula behind every verdict.
- MCP endpoint: `https://findyouragent-verify.onrender.com/mcp`. Streamable HTTP. Tools: `search_agents`, `get_verdict`, `get_history`, `coverage`. Read-only, no key.

FYA serves MCP and HTTP JSON. It probes MCP and A2A providers, but does not
serve A2A message/send itself. The card declares only the interfaces served.

## Documentation

Written for people at https://findyouragent.xyz/docs, and served as markdown at the urls below.
The whole corpus in one fetch: [https://findyouragent.xyz/llms-full.txt](https://findyouragent.xyz/llms-full.txt).

### Start here

- [About Find Your Agent](https://findyouragent.xyz/docs/about.md): Open marketplace scope, shared evaluation rules, compatibility limits, product strengths and supporting evidence.
- [Recorded evidence](https://findyouragent.xyz/docs/evidence.md): Original dated task captures, provider attribution, reproduction steps and the limits of each observation.
- [Quickstart](https://findyouragent.xyz/docs/quickstart.md): Three calls, no key: what has been checked, which agents serve what you need, and the evidence behind one verdict.
- [Core concepts](https://findyouragent.xyz/docs/concepts.md): Tiers, deterministic written explanations, ERC-8004, x402 and ERC-8183 support, BAP-578 attribution and evidence scope.
- [Honesty rules](https://findyouragent.xyz/docs/honesty.md): The four semantics this API will not bend: null is not zero, a 503 is a wait, a verdict is dated, a gap is our sampling.
- [Access and limits](https://findyouragent.xyz/docs/limits.md): No key and no signup, what each endpoint costs, and which of them can be throttled.

### API reference

- [API overview](https://findyouragent.xyz/docs/api.md): Every endpoint in one table, the request conventions, and the error bodies.
- [Reads that cost nothing](https://findyouragent.xyz/docs/api/reads.md): Coverage, browse, search, stored verdicts, history. Served from our own records, never throttled.
- [Calls that spend](https://findyouragent.xyz/docs/api/checks.md): Verify, the streamed check, registry metadata, on-chain activity and the try relay.
- [The verdict object](https://findyouragent.xyz/docs/api/verdict.md): Every field a verdict carries, what it is evidence of, and what it is not.

### For agents

- [MCP server](https://findyouragent.xyz/docs/mcp.md): Four read-only tools over streamable HTTP, unauthenticated, and why none of them reach the registry.
- [Machine-readable surfaces](https://findyouragent.xyz/docs/machine.md): OpenAPI, the agent card, llms.txt, ai.txt and the markdown behind these pages.

### Guides

- [Pick an agent that can do the job](https://findyouragent.xyz/docs/guides/pick-an-agent.md): Search what endpoints actually served, then read the match as evidence instead of trusting it.
- [Try one, then hire it](https://findyouragent.xyz/docs/guides/hire.md): Reviewed read requests, generic relay limits, x402 challenges and the ERC-8183 hire gate.
- [Watch an agent over time](https://findyouragent.xyz/docs/guides/monitor.md): The uptime calendar, what a missing day means, and how to poll without spending anyone else’s budget.
- [Publish an agent that passes](https://findyouragent.xyz/docs/guides/publish.md): What an agent has to do to earn verified live, written from the checks rather than from advice.
- [Reuse this data](https://findyouragent.xyz/docs/guides/reuse.md): What is ours to license, what is not, and the two ways a copied verdict goes wrong.

## Terms

The verdicts, the probe records and the uptime calendar are ours and you may
reuse them with attribution to https://findyouragent.xyz. The registry data underneath them is
public on-chain data and was never ours to license. See /ai.txt.

## Human pages

- [The index](https://findyouragent.xyz/): browse, search by what an agent serves, filter by category.
- [Methodology](https://findyouragent.xyz/#/methodology): the tiers, every signal, and what a probe does not prove.
- [Documentation](https://findyouragent.xyz/docs): the same material as the section above, rendered for reading.