Sources & original observations
MCP statistics and integration research 2026
What the published ecosystem numbers measure, and what one small booking integration actually exposes.
Industry statistics: publisher-reported figures
These figures come from the MCP maintainers. Downloads and registry entries measure different things; neither is a count of paying customers or unique active users.
Approximately 500 million SDK downloads per month
The MCP maintainers reported roughly 500 million monthly downloads across Tier 1 SDKs in their 28 July 2026 release announcement. The announcement describes activity around that release but does not disclose an exact measurement window or counting methodology. Source: The 2026-07-28 Specification, published 28 July 2026.
TypeScript and Python each passed one billion cumulative SDK downloads
The same July 2026 announcement reports that both SDKs crossed one billion total downloads each. These are cumulative package downloads, not one billion developers or installations; the announcement does not specify the cumulative start date. Source: MCP maintainers’ July 2026 announcement.
Nearly 2,000 registry entries in November 2025
The maintainers’ 25 November 2025 anniversary post reported nearly 2,000 MCP Registry entries and 407% growth from the initial September 2025 batch. This is a historical registry snapshot, not today’s server count. The post does not provide a full counting or deduplication method. Source: One Year of MCP, published 25 November 2025.
For a business planning an integration, these figures show ecosystem distribution activity. They cannot establish demand for your particular workflow. Start with the customer task and the integration scope.
Original research: the tools in our booking catalog
Question: how many tools does our current booking workflow expose, and which require authentication?
On 30 September 2026, one complete catalog from our production endpoint contained 6 tool definitions: 2 public tools and 4 tools declaring OAuth authentication. The sample is one server and one catalog retrieval.
Of those six definitions, 3 declare readOnlyHint: true and 3 declare readOnlyHint: false. These are advertised annotations, rather than independent proof of runtime behavior.
| Tool | Declared access | Read-only hint |
|---|---|---|
get_booking_info | Public | true |
get_available_slots | Public | true |
prepare_booking | OAuth | false |
request_booking_verification | OAuth | false |
confirm_booking | OAuth | false |
get_booking_status | OAuth | true |
In this catalog, availability and service information are public. Preparation, email verification, booking confirmation and request status declare authentication. This illustrates the scope of this implementation; it is not a prescribed tool count for every booking product.
Download the measured catalog and calculation inputs (JSON) · Read the connection and booking workflow
Method, reproduction and limits
Measured at , using unauthenticated initialize and tools/list requests to https://mcp-tool.com/mcp. The negotiated protocol version was 2025-11-25. The response contained no next-page cursor, so all returned definitions were included.
We counted each returned tool once. Public tools declare a noauth security scheme; OAuth tools declare oauth2. Read-only counts use the exact boolean value of annotations.readOnlyHint. Missing access schemes or read-only annotations cause the measurement script to fail instead of silently excluding a record.
The download preserves the observed names, annotations and access schemes, together with the timestamp, endpoint and counts. Repeat the two metadata requests and apply those rules to reproduce the measurement. Tool catalogs can change after this snapshot.
Limits: this is a descriptive case study of our own server, with no independent sampling and no control group. We did not execute a tool, send a verification email or create a booking in this measurement. It does not measure email deliverability, completed bookings, conversion, latency or authentication enforcement. No customer records, email addresses, phone numbers or credentials were collected.
What this means for SEO and AI search
Useful evidence needs an identifiable source, a clear unit and a method a reader can inspect. We keep figures in selectable text and link each claim to its source or observation.
Google’s guidance says its usual SEO practices also apply to AI Overviews and AI Mode; it does not require special AI text files or a special schema. Accessible pages, internal links and structured data matching the visible content remain relevant. Publication alone cannot guarantee indexing or an AI recommendation. Source: Google Search Central’s AI features guidance, checked 30 September 2026.
Review history
: first edition. Added three publisher-reported ecosystem figures and a six-tool production catalog snapshot. No successful-booking or conversion findings are included.
Review target: 30 December 2026. This is an editorial target, not an automated update. Future measurements will retain their measurement dates and separate files so that historical references remain inspectable.
About MCP Tool and the person responsible for this research →
Which tools does your product need?
Bring your API docs and one customer workflow. We’ll scope the smallest useful integration.
Discuss your integration ↗