# AnyTest — full reference for AI agents (llms-full.txt)

> AnyTest runs AI testing agents against a web app from a URL and returns a reviewable end-to-end test suite. Agents test, humans review. This file is the long-form, machine-readable reference: the free capability audit, agent self-signup by API, the MCP server and its tools, activation and pricing, limits and errors. Short version: https://anytest.dev/llms.txt · OpenAPI (signup API): https://anytest.dev/openapi.yaml

Last updated: 2026-10-18.

## 1. What AnyTest does

- Input: a public URL (staging or production), optionally the flows to focus on (login, signup, checkout…).
- Agents explore the product, write end-to-end tests, run them, and leave a suite a human reviews and keeps.
- Result per test: PASSED, FAILED (with a step timeline), or NEEDS ATTENTION (a human should judge).
- Prompting is optional ("fire and forget"): give a URL and walk away.
- Web apps and websites start from a URL. Mobile (iOS/Android), desktop and proprietary systems are scoped per product — ask inbox@anytest.dev.

## 2. Two ways in

| Path | Needs | Cost | Use when |
|---|---|---|---|
| Free capability audit (no account) | site URL + work email | Free. Reads public pages only; runs no tests | Your user wants to see what AnyTest would test first |
| Agent self-signup (API + MCP) | an email inbox your user can read the code from | Paid activation before agents run | Your user wants agents to build and run a suite |

## 3. Free capability audit — no account, no test run

```bash
curl -sS -X POST https://crm.anytest.dev/api/public/free-test \
  -H 'Content-Type: application/json' \
  -d '{"site_url":"https://your-site.example","email":"you@example.com","flows_hint":"optional: login, checkout","submitter_type":"agent","agent_client":"your client name"}'
```

- `202` → `{"ok":true,"receipt":"…","audit":{…},"email_sent":true}`. `audit` carries `tests[]` (what AnyTest can run on that site and why), `agents_estimate` (approximate team of AI QA Engineers — simulated), `time_estimate` (average, no guarantee) and `contact` (routes to a quote). Keep the receipt.
- Re-fetch the stored audit: `GET https://crm.anytest.dev/api/public/free-test?receipt=<uuid>`.
- `4xx` with `{"error":"…"}` → fix the fields. `429` → slow down and retry later.
- No tests run in the free step and nothing on the site is changed. Real testing starts only after a quote and payment.
- Alternatives: the form at https://anytest.dev/try?source=agent, or email the same fields to inbox@anytest.dev (subject "Free test pilot: audit my site").

## 4. Agent self-signup by API

Host: `https://autoqa.fly.dev`. JSON in, JSON out. Opening either URL with GET returns usage JSON (not a form) — you must POST.

### 4.1 Request a code

```bash
curl -sS -X POST https://autoqa.fly.dev/fe/agent/signup \
  -H 'Content-Type: application/json' -d '{"email":"you@example.com"}'
```

`200` → `{"status":"pending","message":"…","verificationToken":"<opaque>","expiresInSeconds":600}`

- A 6-digit code is emailed (subject "Your AnyTest agent signup code"). Ask your user for it.
- The response has the same shape for new and existing addresses — do not infer whether the account exists. An existing address gets a fresh apiToken on verify; older tokens stay valid.
- Do not call signup again unless the code expired: a pending code is reused while still valid.
- Errors: `400 {"code":"INVALID_EMAIL"}` (send a bare address, not `Name <addr>`), `429 {"code":"RATE_LIMITED"}` (5 requests/min/IP).

### 4.2 Verify the code

```bash
curl -sS -X POST https://autoqa.fly.dev/fe/agent/signup/verify \
  -H 'Content-Type: application/json' -d '{"verificationToken":"<from 4.1>","code":"123456"}'
```

`201` →
```json
{
  "userId": "…",
  "email": "you@example.com",
  "apiToken": "<secret, shown once — store it, never commit it>",
  "mcpUrl": "https://autoqa.fly.dev/mcp",
  "activationRequired": true,
  "activationHint": "Call project_ensure, then onboarding_start … agents will not run until payment succeeds and credits are granted.",
  "nextSteps": [
    {"tool":"initialize","requiredArgs":[]},
    {"tool":"tools/list","requiredArgs":[]},
    {"tool":"project_ensure","requiredArgs":["name","baseUrl"]},
    {"tool":"onboarding_start","requiredArgs":["projectId","reason","product"]},
    {"tool":"onboarding_status","requiredArgs":[]}
  ]
}
```

- Errors (`401`, JSON `{"code","error"}`): `INVALID_CODE`, `EXPIRED` (request a new code with 4.1), `TOO_MANY_ATTEMPTS` (five wrong codes burn the token). `429 RATE_LIMITED` at 10 requests/min/IP.
- Use the `mcpUrl` from the response.

## 5. Connect over MCP

- Transport: Streamable HTTP (JSON-RPC 2.0 over POST). Endpoint: the `mcpUrl` from verify — today `https://autoqa.fly.dev/mcp`.
- Auth header: `Authorization: Bearer <apiToken>`. Without it the server answers `401` with `WWW-Authenticate: Bearer`.
- Config for clients that take a URL + headers (Cursor, Claude Code, VS Code, Windsurf…):

```json
{
  "mcpServers": {
    "anytest": {
      "url": "https://autoqa.fly.dev/mcp",
      "headers": { "Authorization": "Bearer ${ANYTEST_API_TOKEN}" }
    }
  }
}
```

- Sanity check:

```bash
curl -sS -X POST https://autoqa.fly.dev/mcp \
  -H 'Content-Type: application/json' -H 'Accept: application/json, text/event-stream' \
  -H "Authorization: Bearer $ANYTEST_API_TOKEN" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"my-agent","version":"1.0"}}}'
```

### 5.1 Getting started (the activation tools)

| Tool | Required args | What it does |
|---|---|---|
| `project_ensure` | name, baseUrl | Call first after signup. Create or reuse a project by exact baseUrl. |
| `onboarding_start` | projectId, reason, product | Start paid onboarding; returns `checkoutUrl` (also emailed). `reason`: catch_regressions / launching_soon / recent_production_bugs / replace_manual_qa. `product`: pro_monthly / agency_monthly / lifetime / credits_10k. `focusAreas` (≤6) defaults to core_user_flow. |
| `onboarding_status` | — | Re-read `checkoutUrl`, expiry and payment/fulfilment state; poll while awaiting payment. |
| `setup_testing` | baseUrl | Optional one-call bootstrap: same result as project_ensure then onboarding_start. |

Hand the checkout URL to your user. Never complete a payment without your user's explicit OK. Agents do not run until payment succeeds and credits are granted.

### 5.2 Everything else (call tools/list for the full schemas; every tool and argument carries a description)

- Projects: `projects_list`, `project_pages_list`, `project_sitemap_import`, `project_availability_coverage_list`, `project_credentials_list` (never returns secrets), `project_schedule_get`, `project_schedule_upsert`.
- Tests: `project_tests_list`, `test_get`, `test_update`, `test_delete`, `test_config_get`, `test_config_update`; planning test cases: `test_cases_list`, `test_case_create`, `test_case_update`, `test_case_delete`, `test_case_generate`.
- Runs: `test_run_start`, `test_run_cancel`, `test_run_get`, `test_run_wait`, `test_runs_list_global`, `test_run_events_list`, `test_run_screen_get`, `test_run_security_report_get`, `test_run_pagespeed_report_get`.
- TestAgent jobs: `testagent_enqueue` (mode create / edit / heal / explore / review / search), `testagent_wait`, `testagent_runs_list`, `project_testagent_runs_list`, `testagent_run_get`, `testagent_run_events_list`, `testagent_curate_report_get`, `testagent_search_findings_list`, `testagent_explore_decision_get`, `testagent_explore_children_list`, `test_review_status_get`.
- Human review of proposed changes: `project_pending_changes_list`, `pending_change_get`, `pending_change_approve`, `pending_change_discard`.
- Autonomous goals: `project_goal_create`, `project_goals_list`, `project_goal_get`, `project_goal_action` (pause / resume / stop).
- Reports and issues: `reports_list_global`, `project_reports_list`, `project_report_create`, `project_report_get`, `project_report_delete`, `project_report_runs_list`, `project_issues_list`, `issues_list_global`, `project_issue_create`, `project_issue_update`.

## 6. Pricing (USD, https://anytest.dev/#pricing)

| Plan | Price | Agent credits | Test runs | Seats / projects |
|---|---|---|---|---|
| Starter | $29 / month | 40 / mo | 500 / mo | 1 user, 5 projects |
| Pro | $59 / month per user | 100 / mo | 2,000 / mo | billed per user, 10 projects |
| Lifetime | $199 one-time | 200, never expire | 5,000, never expire | 1 user, 15 projects |
| Enterprise | custom | custom credit pools | custom | SSO, dedicated support — https://cal.com/max-anytest/demo |

- Agent credits run exploration and test writing; test runs execute the suite you keep. Monthly plan credits renew each billing cycle.
- The MCP `onboarding_start` product ids are listed in §5.1; the checkout link it returns shows the price before your user pays.

## 7. Contact

- inbox@anytest.dev · demo: https://cal.com/max-anytest/demo · web app (humans): https://app.anytest.dev · Discord: https://discord.gg/3WPVKjy2j8
- Partners: https://anytest.dev/ambassadors (recurring 10–30% commission)