Do remote MCP servers actually work? (Nov 2026)
56.7%work without auth
33.4%alive but need auth
9.9%failed
293 msmedian initialize latency
7.0median tools per working server
45,311URLs checked
Key facts
- As of 2026-11-19, 56.7% of the 45,311 remote servers in the official MCP Registry checked on 2026-11-19 (25,702) completed an MCP initialize handshake without authentication.
- As of 2026-11-19, a further 33.4% of the 45,311 remote servers in the official MCP Registry checked on 2026-11-19 (15,123) answered but require authentication (HTTP 401/403 or WWW-Authenticate), so 90.1% are alive in total.
- As of 2026-11-19, 9.9% of the 45,311 remote servers in the official MCP Registry checked on 2026-11-19 (4,486) failed with an HTTP error, timeout, DNS, TLS, connection or invalid-response error on a single attempt.
- As of 2026-11-19, among the 25,702 servers that answered initialize without authentication, median initialize latency was 293 ms and p90 751 ms.
- As of 2026-11-19, among the 24,916 working servers that returned a tools/list, the median server exposes 7.0 tools (mean 16.1).
Cite this: Agent Web Census, “Do remote MCP servers actually work?”, https://agents.datacensus.org/mcp/health/, check of remote servers in the official MCP Registry checked on 2026-11-19.
Definitions
- Working: the first remote URL answered one MCP
initializerequest with a valid result and no authentication was demanded. - Auth required: HTTP 401/403 or a
WWW-Authenticateheader. The endpoint is alive but needs credentials; we do not know whether it works for authorised users. - Failed: HTTP error (other 4xx/5xx), timeout (10 s), DNS, TLS or connection error, or a response that is not an MCP initialize result.
- Legacy SSE-type remotes only get a reachability check (event stream starts within 5 s); they have no tool count or protocol version.
Outcome of the handshake
| Outcome | Servers | % of checked |
|---|---|---|
| Working (initialize OK, no auth) | 25,702 | 56.7% |
| Auth required (alive) | 15,123 | 33.4% |
| HTTP error | 2,758 | 6.1% |
| Timeout | 92 | 0.2% |
| DNS error | 9 | 0.0% |
| TLS error | 473 | 1.0% |
| Connection error | 1,051 | 2.3% |
| Invalid response | 103 | 0.2% |
Tools per working server
Latency
Initialize round trip for 25,702 working servers: median 293 ms, p90 751 ms, measured from one network location including connection setup.
Protocol versions
| protocolVersion returned | Servers |
|---|---|
| 2025-06-18 | 21,487 |
| 2025-03-26 | 2,949 |
| 2024-11-05 | 656 |
| 2025-11-25 | 184 |
| 2026-07-28 | 32 |
| 2025-11-05 | 2 |
Biggest hosts
| Host | Servers checked | Working (no auth) |
|---|---|---|
| agent-observatory-sensor.nolimit-observatory.workers.dev | 2,365 | 99.9% |
| gateway.pipeworx.io | 1,936 | 96.5% |
| mcp.apify.com | 1,261 | 0.6% |
| api.mcp.ai | 1,122 | 100.0% |
| cotlet.dev | 398 | 0.0% |
| api.m2mcent.com | 276 | 100.0% |
| server.smithery.ai | 215 | 0.0% |
| atoms.pipavia.com | 145 | 0.0% |
| a2awire.com | 135 | 99.3% |
| textforagents.com | 131 | 75.6% |
| modevc.com | 88 | 53.4% |
| mcp.zovo.one | 82 | 100.0% |
| ukdata.reliness.co.uk | 65 | 100.0% |
| tooloracle.io | 57 | 93.0% |
| api.lazy-mac.com | 54 | 96.3% |
| nova.civai.co | 52 | 100.0% |
| mcp.jojapi.com | 49 | 95.9% |
| thecompanyatlas.com | 48 | 0.0% |
| bumiatlas.com | 45 | 0.0% |
| api.meok.ai | 39 | 0.0% |