Cloud and hosting APIs and MCP servers for AI agents

Cloud providers, serverless and edge platforms, and site hosting an agent can deploy to and manage.

10 cloud and hosting APIs an AI agent can call: auth method and header, official MCP server (10), free tier (9), docs links. Machine-readable JSON included. Agents: fetch catalog/cloud-hosting.json instead of parsing this page. Last checked 2026-09-24.

Ratings and usage (★) fetched: 2026-09-24 16:54 UTC – 2026-09-24 17:02 UTC.

Service	What an agent can do	Auth	MCP server	Free tier	Details (with sources)	Notes	Proof (ratings, usage)
AWS APIs	An agent can provision and control thousands of AWS resources (compute, storage, databases, ML) via SDKs/CLI/REST.	cloud IAM `AWS Signature Version 4 (SigV4) signed request using IAM access key/secret or STS temporary credentials`	official	Free tier: 30+ services always free within monthly limits; new accounts also get up to $200 one-time credits ($100 on signup + up to $100 earned) usable over 6 months	Free plan: 30+ services always free within monthly limits; new customers get up to $200 credits ($100 at signup + up to $100 earned) over 6 months ↗ Example call: verify credentials (STS GetCallerIdentity): POST https://sts.amazonaws.com/ ↗ MCP tools: 8: aws___run_script, aws___get_presigned_url, aws___get_tasks, aws___search_documentation, aws___read_documentation, aws___retrieve_skill … ↗	Umbrella platform: hundreds of distinct service APIs, each with its own IAM scoping, rate limits, and regional availability; no unified OpenAPI spec. The managed AWS MCP Server (OAuth or SigV4 proxy) replaces the local aws-api-mcp-server.	GitHub 9,728★ awslabs/mcp (2026-09-24 16:54 UTC)
Azure APIs	An agent can deploy and manage Azure resources (compute, storage, databases, AI) across resource-provider REST APIs.	cloud IAM `Authorization: Bearer <access_token> (Microsoft Entra ID / Azure AD OAuth2)`	official	12 months of select free services plus $200 credit for 30 days for new accounts; many services also have a permanent always-free monthly amount	Base URL: https://management.azure.com Free plan: 65+ always-free services with monthly amounts; new customers also get 20+ services free for 12 months and $200 credit for 30 days, card required ↗ Rate limits: ARM token bucket per subscription and per tenant: reads 250 bucket/25 per s refill; writes and deletes 200 bucket/10 per s; 429 with Retry-After ↗ Async jobs: some operations are asynchronous; track them via response headers (Track asynchronous Azure operations) ↗ Example call: create or update a resource group: PUT https://management.azure.com/subscriptions/{subscriptionId}/resourcegroups/{resourceGroupName}?api-version=2021-04-01 ↗	Umbrella platform: hundreds of resource-provider APIs, each independently versioned (api-version query param required); the Azure MCP repo moved from Azure/azure-mcp (archived) to microsoft/mcp.	GitHub 3,094★ Azure/azure-rest-api-specs (2026-09-24 16:54 UTC) GitHub 3,707★ microsoft/mcp (2026-09-24 16:54 UTC)
Cloudflare API	An agent can manage DNS, Workers, zones, WAF/security rules, and caching for Cloudflare accounts programmatically.	API key / OAuth `Authorization: Bearer <API_Token>`	official	free with any Cloudflare account; API access itself is free, manages both free and paid Cloudflare products	Base URL: https://api.cloudflare.com/client/v4 Free plan: Free plan ($0); API access included ↗ Rate limits: 1,200 requests per 5 minutes per user/account token; 200 req/s per IP; GraphQL max 320/5 min by query cost; exceeding blocks calls for 5 min (429) ↗ Example call: create a DNS record: POST https://api.cloudflare.com/client/v4/zones/{zone_id}/dns_records ↗ MCP tools: 2: search, execute ↗	Global rate limit ~1,200 requests/5 min per user; per-IP limit ~200 req/sec; GraphQL Analytics API has a separate cost-based limit. Legacy X-Auth-Key global key auth also exists but is discouraged.	Smithery 199 uses (2026-09-24 16:54 UTC) SourceForge 4.6★/12 reviews (2026-09-24 16:57 UTC)
Fly.io Machines API	An agent can create, start, stop, and destroy Fly.io Machines (lightweight VMs) programmatically via REST.	API key `Authorization: Bearer <FLY_API_TOKEN>`	official	Trial: 2 machine-hours or 7 days, whichever comes first, no card needed to start (adding a card ends the trial); then pay-as-you-go, no perpetual free tier	Base URL: https://api.machines.dev/v1 Free plan: 2 hours of machine runtime or 7 days, whichever comes first (up to 10 machines, 20GB volumes), one-time, no card ↗ Rate limits: 1 req/s per action (e.g. create, start) with burst to 3 req/s; Get Machine 5 req/s, burst 10 ↗ Async jobs: machine state changes are asynchronous; block with GET /v1/apps/{app}/machines/{id}/wait?state=started (default timeout 60s) ↗ Example call: create and start a Machine (VM): POST https://api.machines.dev/v1/apps/{app_name}/machines ↗ Unit price: shared-cpu-1x 256MB Machine: $0.00000078 per second running (~$0.0028/hour) base price, regional multipliers apply; stopped rootfs $0.15 per GB per 30 days ↗ MCP tools: 9: apps, certs, logs, machine, orgs, platform … ↗	The official MCP server for Machines is the local flyctl server ('fly mcp server'); the hosted Sprites MCP controls Sprites, not Machines.	—
Google Cloud APIs	An agent can create and manage Google Cloud resources (compute, storage, data, AI/ML) via REST, gRPC, or client libraries.	cloud IAM `Authorization: Bearer <OAuth2 access token> (service account or Application Default Credentials); API keys for select public APIs`	official	free trial credit for new customers plus an Always Free tier with perpetual no-cost limits on select products	Free plan: Free Tier: monthly always-free limits on 20+ products; plus $300 Welcome credit over 90 days for new customers, card required ↗ Example call: get a project (Resource Manager): GET https://cloudresourcemanager.googleapis.com/v3/projects/{PROJECT_ID} ↗	Umbrella platform: hundreds of distinct service APIs, each with its own scopes/quotas; many APIs require explicit per-project enabling. Google runs official remote MCP servers per product (e.g. compute.googleapis.com/mcp, run.googleapis.com/mcp, bigquery.googleapis.com/mcp).	GitHub 16.5k★ googleapis/mcp-toolbox (2026-09-24 16:54 UTC)
Mux Video API	Upload, encode, stream and analyse video; get playback URLs, thumbnails and captions via REST.	API key `HTTP Basic auth: <token_id>:<token_secret>`	official	first 100,000 delivered minutes/month free, then usage-based	Base URL: https://api.mux.com Free plan: 100,000 delivery minutes, per month, no card ↗ Rate limits: Per-environment rate-limit buckets; HTTP 429 when exceeded ↗ Example call: create a video asset from a URL: POST https://api.mux.com/video/v1/assets ↗ Unit price: Basic-quality input free; storage from $0.0024 per minute stored per month; delivery 100,000 min/month free, then from $0.0008 per minute delivered ↗ MCP tools: 2: docs search, code execution (TypeScript against @mux/mux-node) ↗	Hosting/streaming infrastructure; does not generate video from prompts.	—
Netlify API	An agent can create/manage sites, trigger and monitor deploys, and configure DNS, domains, and env vars.	API key / OAuth `Authorization: Bearer <PERSONAL_ACCESS_TOKEN>`	official	Free: $0 forever plan with 300 credits per month (hard limit, no rollover); API usage subject to account plan limits	Base URL: https://api.netlify.com/api/v1 Free plan: Free plan, $0 forever, 300 credit limit ↗ Rate limits: 500 requests/min general; deploys 3/min and 100/day; X-RateLimit-* headers ↗ Example call: deploy a site from a ZIP file: POST https://api.netlify.com/api/v1/sites/{site_id}/deploys ↗ Unit price: Production deploy costs 15 credits; add-on credits $5 per 500 (Personal, ~$0.15/deploy) or $10 per 1,500 (Pro, ~$0.10/deploy) ↗	Rate limits: 500 req/min general, 3 deploys/min (100/day). Tokens are invalidated on password reset.	SourceForge 4.0★/5 reviews (2026-09-24 16:57 UTC)
Render API	An agent can create and manage Render services, trigger deployments, and configure databases and env vars.	API key `Authorization: Bearer <api-key>`	official	free static sites; free web services that spin down after inactivity	Base URL: https://api.render.com/v1 Free plan: 750 free instance hours per workspace per month; free web services spin down after 15 min idle; free Postgres expires 30 days after creation ↗ Example call: trigger a deploy: POST https://api.render.com/v1/services/{serviceId}/deploys ↗ MCP tools: 25: list_workspaces, select_workspace, get_selected_workspace, list_services, get_service, create_web_service … ↗	OpenAPI spec endpoint/tag names may change over time even though the API itself stays backward compatible. Render's official MCP server supports potentially destructive operations (env var changes, triggering deploys).	—
Tencent EdgeOne Pages (Makers)	Lets an agent deploy HTML pages, static sites and full-stack apps with edge functions to EdgeOne and get back a public URL.	API key `API token created in the EdgeOne Makers console (API Token tab), passed to the MCP server or CLI as EDGEONE_PAGES_API_TOKEN; the hosted HTML-sharing MCP needs no token`	official	Free plan (stated as permanent): 40 projects, 500 builds/month, 3M Edge Function and 1M Cloud Function invocations/month, 5 GB storage, 1 GB KV, 1 GB Blob.	Free plan: free_tier ↗ MCP tools: 1: deploy-html ↗	Renamed EdgeOne Makers. No public REST API: use the MCP server (@edgeone/makers-mcp) or EdgeOne CLI. The remote endpoint's no-login tool deploys a single HTML page; projects need the token.	—
Vercel API	An agent can create deployments, manage projects/domains/env vars, and query analytics and logs.	API key / OAuth `Authorization: Bearer <access_token>`	official	free on Hobby plan; some resources/endpoints require Pro/paid plan	Base URL: https://api.vercel.com Free plan: Hobby plan $0/mo with capped monthly allotment per billable resource; 100 deployments/day ↗ Rate limits: Per-endpoint limits, e.g. deployments list 1,000/min, env var reads 500/min, Hobby deployments 100/day and 100/hour; 429 on excess ↗ Example call: create a deployment from inline files: POST https://api.vercel.com/v13/deployments ↗ MCP tools: 20: search_vercel_documentation, list_teams, list_projects, get_project, list_deployments, get_deployment … ↗	Vercel MCP (Beta) is remote/OAuth-based; write actions like deployments or purchases require explicit OAuth consent.	Smithery 549 uses (2026-09-24 16:54 UTC) SourceForge 4.5★/2 reviews (2026-09-24 16:57 UTC)

Cloud and hosting APIs and MCP servers for AI agents

Other categories