State of MCP Readiness

An independent, continuously re-run, cryptographically signed measurement of the Model Context Protocol ecosystem. Measured from outside — not an endorsement, not a safety scan. Generated 2026-07-01 · Standard agent-tool-discoverability-standard/0.2

Of 15,121 MCP endpoints observed, 19.1% answer an anonymous MCP handshake. Among those that respond, 88.9% are observed-ready (grade A/B) and 56.4% returned verifiable content to an anonymous probe.

Liveness across the ecosystem

MCP endpoints observed

15,121

Answer an anonymous handshake

19.1%

2,887 servers

No anonymous handshake

80.9%

12,234 (dead / moved / auth-gated)

"Did not complete a handshake" means the listed URL did not answer an anonymous MCP initialize at measurement time — dead, moved, mis-hosted, not actually an MCP endpoint, OR requiring credentials even to initialize. It is a liveness fact, NOT a claim that a working server is broken, and NOT a death certificate.

Among the 2,887 servers that respond

Observed-ready (A / B)

88.9%

Returned verifiable content

56.4%

Grade A

424

Grade	A	B	C	D
servers	424	2143	304	16

The remainder is UNVERIFIED, not 'ghost'. Per SaSame's own finding 0009, the majority of empty results to an anonymous synthetic probe are a probe limitation (a required-arg / no-match / validation response, or a tool that faithfully rendered our trivial input) — NOT evidence the tool is hollow.

Most common gaps (among responding servers)

Criterion	What it checks	Servers missing it
`C7`	returns verifiable content	43.6% (1258)
`C5`	safety annotations	39.8% (1148)
`C9`	token efficiency	35.9% (1037)
`C10`	honest error behavior	7.6% (218)
`C3`	tool-object validity	5.6% (161)
`C4`	description sufficiency	5.5% (160)

Change over time (servers observed 2+ times: 2,612)

Stable	2227
Improving	6
Substantive degraders (lost ≥2 criteria)	5
Measurement-limited changes (single-criterion / likely transient)	374
Stopped responding since first seen	2

Degraders are split per finding 0006: SUBSTANTIVE = lost >=2 criteria including a non-C7 one; MEASUREMENT-LIMITED = single-criterion or C7-only change that is likely transient (finding 0009). Covers only servers observed 2+ times; a lower bound on real churn.

Observed-ready leaderboard (grade A — a sample of 424)

Server	Score
ai.agentrapay/agentra	10/10
ai.airshelf/catalog	10/10
ai.atdev/supershopping	10/10
ai.autonomad/travel	10/10
ai.baselight/baselight	10/10
ai.childadhd/library	10/10
ai.childanxiety/library	10/10
ai.childpsychiatry/library	10/10
ai.compeller/compel	10/10
ai.demanddiscovery/mcp	10/10
ai.exa/exa	10/10
ai.ironscout/ironscout	10/10
ai.keenable/web-search	10/10
ai.keenable/web-search	10/10
ai.modulos/demo-booking	10/10
ai.pocketdrives/pocket-drives	10/10
ai.quantifyme/quantifyme	10/10
ai.readypermit/geo	10/10
ai.rfix/mcp	10/10
ai.searchshop.www/la-luer	10/10
ai.teenadhd/library	10/10
ai.teenanxiety/library	10/10
ai.teenpsychiatry/library	10/10
ai.villiers/charter	10/10
ai.weftly/weftly	9/10

Only grade-A servers are named. Lower grades are reported only in aggregate — we publish a positive leaderboard, never a wall of shame.

On curated directories

Curated, human-reviewed directories (e.g. Anthropic's Connectors Directory of ~511 vetted connectors) publish company homepages, not MCP endpoints, and are ~97% OAuth-gated — so they are not externally verifiable and not continuously re-checked. That is precisely the gap an independent, continuous, machine-verifiable measurement layer fills. This report COMPLEMENTS point-in-time human review; it does not contradict it.

Verify this report yourself

This report is ed25519-signed by SaSame's issuer key and offline-verifiable with no callback:

ed25519.verify(base64→bytes(signature), utf8(canonical_json), issuer_pubkey_spki_hex)

Issuer pubkey: 302a300506032b6570032100439ce47d… · Signed JSON: state-of-mcp-report.json

SaSame MCP Observatory — a neutral, third-party readiness & trust measurement layer for the AI agent economy. Grades are observations, never endorsements or safety/malware verdicts. Method: 10 falsifiable criteria (agent-tool-discoverability-standard/0.2). Corrections & delistings: consulting@srl-sasame.com.