Your AI now knows every data center.
datacenters-world is a hosted MCP server over the largest open atlas of data center facilities on the internet. Add one config snippet and Claude Desktop, Cursor, or Claude Code can answer infrastructure questions about 5,675 facilities, 34,732 networks, and 1,309 Internet exchanges — with a citation URL on every result.
Why
Ask Claude “which data centers does Equinix run in Frankfurt” today and you get a memorized list with no citations, no power numbers, no ASN counts. The model doesn’t have structured data — and the web pages it can browse are slow, half-paywalled, and inconsistent across operators.
This MCP server hands the model the actual dataset: facility records, operator-published specs, peering memberships from PeeringDB, IXP rosters, and hyperscale cloud regions. Every tool response carries a source_url pointing back at the canonical page on datacenters.world, so the model can cite. Free tier covers ~150–300 chat conversations per month.
What the agent can answer
Drop into any MCP-aware client and ask, in natural language:
- Ask
Compare Equinix vs Digital Realty in Frankfurt — facility count, total power, ASNs peered, IXPs reachable.
- Ask
List every Tier IV data center in Singapore with at least 10 MW and BGP peering at SGIX.
- Ask
For each AWS region in Europe, name the city, country, and the nearest carrier-neutral facility with 100+ ASNs.
- Ask
Which operators have facilities in more than 20 countries? Rank them by total network count.
Install
One snippet per client. The endpoint is streamable HTTP, so any modern MCP-aware tool can talk to it directly. The bearer key gates access and tracks usage — create one for free from the dashboard.
{
"mcpServers": {
"datacenters-world": {
"url": "https://datacenters.world/api/mcp",
"headers": {
"Authorization": "Bearer dcw_…"
}
}
}
}claude mcp add datacenters-world \ --transport http \ --header "Authorization: Bearer dcw_…" \ https://datacenters.world/api/mcp
mcp-remote{
"mcpServers": {
"datacenters-world": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://datacenters.world/api/mcp",
"--header",
"Authorization: Bearer dcw_…"
]
}
}
}Tools
Five tools, each typed with Zod schemas so the model picks the right one without guessing.
| Tool | Inputs | Returns |
|---|---|---|
| search_facilities | country[], operator (prefix), min_power_mw, limit | Up to 50 facilities with key facts + source_url. |
| get_facility | slug | Full record — specs, top 50 ASNs, top 20 IXPs, source provenance. |
| list_operators | country[], min_facilities, limit | Operators ranked by facility count + country breadth. |
| list_countries | (none) | All 148 countries with facility counts. |
| list_cloud_regions | provider, country[] | Hyperscale regions across AWS, Google, Azure, Oracle. |
Pricing
Same quota as the REST API — one key works for both surfaces. Protocol overhead (initialize, tool list, notifications) is not charged; only tools/call counts against your monthly quota.
| Tier | Tool calls / month | Price |
|---|---|---|
| Free | 1,000 | $0 |
| Pro | 10,000 | $9.99 / mo |
| Team | 50,000 | $39.99 / mo |
| Enterprise | custom | contact |
Pro and Team include a 3-day free trial. Quotas roll over monthly — anniversary date for Free, billing-period for paid.
Where the data comes from
PeeringDB for interconnect-relevant facilities, operator-published pages (Equinix, Digital Realty, DataBank, Cologix, CoreSite, CyrusOne, QTS, Iron Mountain), OpenStreetMap for location cross-checks, plus dedicated scrapers for datacenters.google and datacenters.atmeta.com. Full methodology, source counts, and known gaps live on /methodology.
The whole dataset is also browsable as a public map at datacenters.world, and the same atlas is exposed as a JSON/CSV REST API at /api.
Plug it in.
Sign in with GitHub, create a key, paste the snippet. Sixty seconds to working tool calls inside Claude or Cursor.