# BotFriendly.xyz

BotFriendly is a directory of services AI agents can actually use.
We list APIs, tools, and merchants with machine-readable interfaces (MCP
servers, OpenAPI specs, llms.txt), machine-completable auth, and support
for agent payment and commerce protocols (x402, MPP, L402, AP2, TAP, Agent Pay, ACE, ACP, UCP, ACTP, ICC).
Capability badges are probe-backed: a badge is only set when we fetched
the endpoint and something answered. Humans optional.

## When to use BotFriendly

Reach for this site when you are an agent (or building one) and need to:

- Find a service for a capability you lack — browser automation, vector
  memory, payments, voice, OCR, code execution, search — filtered to ones
  with machine-usable interfaces (MCP, OpenAPI, API-key or M2M auth).
- Check whether a specific vendor is agent-ready before integrating:
  GET /api/v1/services/{slug} returns its auth method, MCP/OpenAPI URLs,
  and supported payment protocols.
- Find merchants and APIs that accept agent-initiated payments over a
  specific protocol: GET /api/v1/services?protocol=x402 (or acp, ucp, mpp…).
- Score how bot-friendly a site is (yours or a vendor's) before relying
  on it: GET /api/v1/check?url=example.com.

Not the right tool for: general web search, package registries, or model
benchmarks — we only index services with machine-readable interfaces.

## API

Base URL: https://botfriendly.xyz/api/v1

### Endpoints

- GET /api/v1/services — List and filter all services
  Query params: ?category=browse&has_mcp=true&no_human=true&accepts_mpp=true&has_skill_file=true&has_llms_txt=true&auth_method=api-key&search=browser&min_score=4&sort=score&promoted=false&page=1&per_page=20
  Sort options: score (default), votes, name, newest
  Ads: results may start with up to 3 promoted listings (paid ads), always marked promoted:true and ad:true; pass promoted=false to exclude them
- GET /api/v1/services/{slug} — Get a single service by slug (full details)
- GET /api/v1/search?q=... — Full-text search across services
  Query params: ?q=vector+database&category=remember&page=1&per_page=20
- GET /api/v1/categories — Categories with labels, aliases and service counts
  Filter with ?category=<id>; a merged category's old slug (trade, bet, connect, automate, locate, book, sign, hire) still works and selects just that slug
- GET /api/v1/featured — Get site-wide promoted/featured services (max 3)
- GET /api/v1/protocols — Protocol adoption stats, sponsors and spec URLs (MCP, OpenAPI, A2A, Cross App Access, x402, MPP, L402, AP2, TAP, Agent Pay, ACE, ACP, UCP, ACTP, ICC); A2A and Cross App Access aren't listing attributes, so their count is null. Human-readable: https://botfriendly.xyz/protocols. Agent SDKs and frameworks (not listings): https://botfriendly.xyz/libraries
- GET /api/v1/services?protocol=x402 — Filter listings by payment protocol (repeat the param to require several)
- GET /api/v1/stats — Directory stats (total services, per-category counts, protocol coverage)
- POST /api/v1/verify — Probe a service's claimed interfaces and return the verdicts (no auth required, nothing written)
  Body: {"url":"https://example.com","has_mcp_server":true,"mcp_server_url":"https://example.com/mcp"} — the same check applied to submissions and discovery
- POST /api/v1/submissions — Submit a new service (no auth required; published or rejected on the spot by probing its interface URLs)
- GET /api/v1/health — Health check and endpoint discovery
- GET /api/v1/check?url=... — Score how bot-friendly any site is (access, no-JS content, machine discovery), with advice per failed check
  Add &format=prompt to get a ready-to-paste coding-agent prompt that fixes every failed finding, instead of the JSON report
- GET /api/v1/openapi.json — Full OpenAPI 3.1 specification

### Authentication

No authentication required for read endpoints. Rate limited to 100 req/min by IP.

### Machine-Readable Interfaces

- Remote MCP Server (Streamable HTTP): https://mcp.botfriendly.xyz
  Hosted, no key: the URL is the whole install. Add it as a custom connector in Claude, ChatGPT, Cursor, or any MCP client.
  Claude Code: claude mcp add botfriendly --transport http https://mcp.botfriendly.xyz
  Manifest: https://botfriendly.xyz/.well-known/mcp.json
- OpenAPI Spec: https://botfriendly.xyz/api/v1/openapi.json
- AI Plugin Manifest: https://botfriendly.xyz/.well-known/ai-plugin.json
- Markdown content negotiation: request any main page with Accept: text/markdown to get a markdown representation.

### Verified vs claimed interfaces

Every service carries flags like has_mcp_server, has_openapi_spec and has_api. Those are
what the listing claims. The verified_interfaces array on each service is the
subset we have actually fetched and confirmed — an MCP server that answered a
handshake, a spec that parsed, an API host that answered in JSON, an llms.txt
that was really markdown.

no_human_needed (filter with ?no_human=true) marks a service an agent can
start using with nobody setting up an account or a key: its own MCP server
listed its tools to an unauthenticated request and the service needs no key
(no auth, or a wallet the agent holds), or one of its interfaces answered
HTTP 402 asking to be paid per request (x402, MPP, L402). Documentation MCP
servers don't count. It is computed from the probe evidence, never claimed.
An API that wants a key is still fine to use — it just needs a human once.

A flag that is true but missing from verified_interfaces is unverified: nobody
has shown us the artifact yet. It may well exist. Prefer a verified interface
when choosing how to call a service, and treat an unverified one as a lead
rather than an address.

## Categories

  - think: Models — Chat, embedding, reasoning, and other model APIs — not only LLMs
  - see: Vision — OCR, document parsing, and vision products
  - hear-speak: Speech — Speech-to-text, text-to-speech, voice
  - create: Media — Image, video, and music generation
  - remember: Memory — Vector DBs, stores, agent memory
  - search: Search — Web and semantic retrieval, maps, geodata (also accepts: locate)
  - browse: Browsers — Headless browsers, scraping, web automation
  - communicate: Messaging — Email, SMS, chat, notifications
  - execute: Compute — Sandboxes, serverless, GPUs
  - pay: Payments — Rails and facilitators — x402, MPP, L402
  - shop: Merchants — Goods, services, and bookings agents can buy (also accepts: book)
  - markets: Markets — Exchanges, brokerages, prediction and betting markets (also accepts: trade, bet)
  - platforms: Platforms — Agent frameworks, tool platforms, workflow engines (also accepts: connect, automate)
  - identify: Identity — M2M auth, agent identity, e-signature (also accepts: sign)
  - monitor: Observability — Tracing, error tracking, agent ops
  - other: Other — Everything else agents can use, including work marketplaces (also accepts: hire)

## Example Queries

Search for browser automation tools with MCP:
  curl "https://botfriendly.xyz/api/v1/services?category=browse&has_mcp=true"

Find all services accepting x402 payments:
  curl "https://botfriendly.xyz/api/v1/services?accepts_x402=true"

Full-text search for services:
  curl "https://botfriendly.xyz/api/v1/search?q=vector+database"

Get services with a bot-friendly score of 5, sorted by votes:
  curl "https://botfriendly.xyz/api/v1/services?min_score=5&sort=votes"

Find services with skill files or llms.txt:
  curl "https://botfriendly.xyz/api/v1/services?has_skill_file=true"
  curl "https://botfriendly.xyz/api/v1/services?has_llms_txt=true"

Get featured/promoted services:
  curl "https://botfriendly.xyz/api/v1/featured"

Score how bot-friendly a site is:
  curl "https://botfriendly.xyz/api/v1/check?url=yourservice.dev"

## Submitting a Service

POST https://botfriendly.xyz/api/v1/submissions
Content-Type: application/json

{
  "name": "YourService",
  "url": "https://yourservice.dev",
  "category": "browse",
  "description": "What your service does for agents",
  "auth_method": "api-key",
  "has_mcp_server": true,
  "contact_email": "you@example.com"
}

No authentication required. There is no review queue: the interface URLs
you claim are fetched during the request, and the listing publishes
immediately if its MCP server, OpenAPI spec or API completes a handshake
(one of has_mcp_server / has_openapi_spec / has_api is required; llms.txt,
skill files and feeds only add badges). Interface URLs sit on the
service's own domain, or on another host the service's own site links to. The url must be the product, not an article,
and a domain can add at most 2 listings a day and 3 in total.
Descriptions are generated, not taken verbatim.

## Contact

- Email: hi@botfriendly.xyz
- X/Twitter: @BotFriendly_xyz