FOR DEVELOPERS & AI AGENTS

Query the Mac Cloud Index

Give your assistant access to the same sourced catalog you see in the directory: macOS CI runners, dedicated Macs and virtual machines.

Connect with MCP

Add this URL as a remote MCP server in your client. Use Streamable HTTP; no account or API key is required.

https://maccloudindex.com/mcp

A typical client configuration looks like this. Your client may also require a transport field such as "type": "http".

{
  "mcpServers": {
    "mac-cloud-index": {
      "url": "https://maccloudindex.com/mcp"
    }
  }
}

ChatGPT / Claude connectors

In ChatGPT or Claude, add a custom connector / remote MCP server and paste https://maccloudindex.com/mcp. No OAuth or API key. Then ask budget-aware questions such as: “Using Mac Cloud Index MCP, find CI runners under $100 for 1,000 minutes in one month” or “Use preset dedicated-budget and list sources.”

Publishing an MCP endpoint does not auto-connect every chat product — the user (or workspace admin) must add the connector once.

Available tools

list_providers: Find companies or sort by a documented company dimension, separately from offer value.
get_catalog_filters: Discover supported chips, regions, billing units and software versions for a category.
search_offers: Filter configurations by category, provider, chip, RAM, region, billing, software, included service, documented capability, USD maxCost budget or a named scenario preset.
get_provider: Read a company profile, shared evidence, MCI assessment breakdowns, review gaps and its resource listings.
get_offer: Read complete configuration details, native prices, billing minimums, source links and observation dates.

Try a catalog query

Ask your assistant: “Find dedicated M4 Macs with at least 32 GB of RAM. Show native prices, minimum commitments and sources.”

Or use a preset: search_offers with category: "ci" and preset: "ci-budget" ($100 ceiling for one month). Categories are ci, dedicated and vms. Search text matches literal text; the client turns your question into structured filters. maxCost compares USD modeled bills only.

Keep the evidence with the answer

Every data response identifies its catalog release and publication time. Facts include source links and observation dates. Publication time is not the date every fact was checked.

Prices retain their native currency, billing period, minimum commitment and regional scope. The catalog does not establish live stock. Missing information remains unknown, and inferred capabilities require confirmation from the provider. Retrieve complete offer details before comparing prices or making a recommendation.

Company grades and offer services

Offer summaries include canonical services with delivery, entitlement, conditions and source evidence. Built-in publishing is distinct from customer-run upload scripts. Use get_catalog_filters to discover service values; the service filter requires current, explicitly supported and included services.

companyAssessment provides fixed 0–3 grades for SLA, support, security, data lifecycle, workflow and billing clarity. Company grades exclude price, hardware value and identity bonuses. list_providers accepts sort: "sla" (or support, security, data, workflow, billing) and an optional service category. Category scopes Workflow only; the other dimensions remain company-wide. get_provider returns full criterion evidence and reference-plan limits. Every company has all six numeric grades: 0 No evidence, 1 Bad, 2 Good, 3 Excellent. Missing or conflicting required evidence receives 0, with its evidence state preserved; zero is not proof of poor service. There is no overall company score. Company JSON schema version 2 includes these labels and numeric criterion scores.

Company JSON is separate from offer JSON. Both use the same published release and listing policy. Legacy providerAssessment and sort: "mci" remain compatibility-only; use company dimensions for current company rankings and value for offer cost scores. Read the scoring methodology.

Machine discovery

Agents can locate this catalog without scraping HTML:

llms.txt — site summary and preferred entry points
Category markdown mirrors — lean answer shortlists
Cheapest Mac CI, dedicated vs VM billing, and rent a Mac for local AI
MCP server card — Streamable HTTP endpoint metadata
RFC 9727 API catalog — linkset for JSON and MCP
Agent skills index — catalog query skill
Indexable category landings: /runners, /dedicated-macs, /vms

robots.txt allows search indexing and AI input/citation (ai-input=yes) while disallowing model training (ai-train=no).

Usage

Offering summaries include value: a numeric 0–100 score for every offering, the modeled bill or partial estimate, assumptions, evidence and reference prices. The price formula is 100 × reference price ÷ (reference price + offering price). Missing fields broaden the comparison or produce a provisional capacity score; no usable inputs give a neutral 50. Eligible zero-cost usage scores 100. search_offers accepts sort: "value", maxCost, preset, ciMinutes, months, jobMinutes, repository, appleMembership and rentalHours. Shared account credits are assumed already used. Provider assessment fields remain for compatibility and do not contribute to offering value.

Read-only public data. Results are paginated, with up to 20 listings per call. The endpoint applies an approximate limit of 120 POST requests per minute per IP at each Cloudflare location; clients sharing an IP share that allowance. On HTTP 429, respect the Retry-After header.

Native MCP clients can connect without a browser Origin header. The public, keyless endpoint accepts browser Origins; host validation and rate limits still apply. A connection is required: publishing an MCP endpoint does not automatically connect every AI service.

Explore the directory · Browse the public JSON API