MCP adoption is past the noise floor. Q2 2026 closed with 9,400 published servers across the four major registries, sustaining a +58% QoQ growth rate that has held three quarters in a row. The 6-month forecast brackets year-end at 14,800–22,000 servers across three confidence bands, with the base case at 18,400.
What follows is the per-registry projection, the enterprise deployment shape we expect through Q3, the standards-war risk (medium, but not zero), and the specific signals to watch through August. The forecast is most useful as a planning input for integration roadmaps, not as a single point estimate.
- 01Base-case Q3 2026 forecast: 18,400 published servers (+96% from Q2 baseline).Sustained 35–40% QoQ growth across the four registries. Aggressive scenario (0.30 weight): 22,000 if enterprise registries fast-track. Conservative (0.20 weight): 14,800 if registry fragmentation slows growth. Probability-weighted forecast: 18,640.
- 02Curated registries (Glama, Cloudflare AI MCP) grow faster than community registries.Glama projects to 4,200 (Q3) from 2,750 (Q2); Cloudflare AI MCP projects to 1,400 from 620. The growth-by-quality pattern is the maturation signal — enterprise procurement filters increasingly favor curated lists, pulling vendor commitment toward them.
- 03Enterprise pilot-to-production conversion via MCP-integrated stacks projects to 41–47% by Q3.Q2 baseline was 31% conversion overall, with MCP-integrated pilots converting at 38% versus 22% for non-MCP. The MCP-integrated cohort grows; the differential persists. By Q3 we expect overall conversion 41% with MCP cohort at 47%.
- 04Standards-war risk is medium, not high — Anthropic remains primary stewards but extension proposals are accelerating.Microsoft, OpenAI, and Google have shipped MCP-compatible servers and clients but proposed extensions diverge in places. Risk that Q3–Q4 sees a competing 'extended-MCP' track is real but coordinated through committee structures. Watch the standards-track meetings in July.
- 05The dominant Q3 watch-signal is enterprise-vendor first-party server announcements, not raw count growth.Q2 added Atlassian, Salesforce, Stripe, GitHub, Linear. Q3 should add HubSpot, Zendesk, Notion, Slack, ServiceNow, Workday — meaning published-server count growth that's heavy with high-quality enterprise-grade servers. The composition matters more than the count.
01 — The Q2 BaselineWhere adoption stands today.
The Q2 2026 close gave us a clean baseline: 9,400 published servers across the four major registries, with a sustained +58% QoQ growth rate over the prior three quarters. The growth distribution is the interesting part — not all registries grew evenly.
Community-leaning · Q2 close
Largest registry by raw count, dominated by community contributions. Q2 growth +52% QoQ. Quality is variable; sustained-uptime servers are roughly 35% of total. Best for discovery, weaker for production picks.
+52% QoQVendor-curated · Q2 close
Curated catalog with paid-tier support. Q2 growth +71% QoQ — fastest among the four. Production-leaning; quality bar is higher. Used by enterprise procurement teams to short-list MCP integrations.
+71% QoQOpen-source-leaning · Q2 close
Open-source catalog with strong tool-call testing harness. Q2 growth +44% QoQ. Lowest barrier to listing, highest variance in quality. Good for engineering teams that want to evaluate before committing.
+44% QoQManaged-runtime · Q2 close
Cloudflare-hosted servers with managed runtime. Q2 growth +103% QoQ — the largest by percentage. Smallest count, highest reliability. The fastest-growing registry by deploy-count Q2.
+103% QoQ"Curated and managed-runtime registries are growing fastest. Counts matter; composition matters more."— Internal forecast memo, April 2026
02 — Three Q3 ScenariosThe forecast range.
We model three scenarios for the published-server count by end of Q3 2026 (September 30). The base-case scenario carries 50% probability weight, the aggressive scenario 30%, the conservative scenario 20%. Probability-weighted forecast: 18,640 servers.
22,000 servers · enterprise fast-track (0.30 weight)
Curated and managed-runtime registries continue current growth pace; enterprise vendors fast-track first-party server commitments tied to Q3 product launches; Cloudflare AI MCP doubles to 1,400+ on managed-runtime momentum. The standards-war risk does not materialize.
0.30 probability18,400 servers · sustained QoQ growth (0.50 weight)
All four registries sustain 35–45% QoQ growth (slowing modestly from Q2). Enterprise vendor announcements continue at Q2 cadence (~12 per quarter). Standards-war risk remains contained; no major fragmentation event. Most likely path.
0.50 probability14,800 servers · registry fragmentation slows (0.20 weight)
A standards-war event in Q3 (extended-MCP track formalized, OpenAI ships incompatible variant, or major-vendor pulls back) introduces enough fragmentation that registry growth flattens. Smithery still grows on community submission; curated registries slow. Unfavored but credible.
0.20 probabilityMCP published-server count · 5 quarters · Q3 forecast bands
Forecast: weekly registry snapshots · scenario-weighted · Apr 202603 — Per-Registry ForecastThe composition shift.
Aggregate growth obscures the composition shift. The four registries grow at materially different rates and compose toward the curated and managed-runtime end. By Q3, Glama and Cloudflare AI MCP combined will represent ~30% of total servers, up from 36% of Q2 (yes — that's a slight share decrease as Smithery's Q3 community submission wave pulls share back).
9,800 · Q3 base case
Community-leaning · +33% QoQ projectedContinued community submission wave. Q3 community submissions tend to spike around developer conferences and AI-tooling launches. Quality remains variable; sustained-uptime servers grow as a share but stay below 40%.
Largest count4,200 · Q3 base case
Vendor-curated · +53% QoQ projectedFastest-growing registry by raw new servers. Curated catalog admission process prioritizes enterprise-grade servers; Q3 growth driven by HubSpot, Zendesk, Notion, ServiceNow first-party server submissions.
Curated growth3,000 · Q3 base case
Open-source-leaning · +65% QoQ projectedQ3 growth driven by OSS toolkit integrations and developer-tool servers. Maintains lowest barrier to listing; testing-harness coverage expanding to cover more server types. Variance in quality grows with count.
OSS growth1,400 · Q3 base case
Managed-runtime · +126% QoQ projectedHighest growth rate by percentage. Managed-runtime pattern (MCP-over-Workers) attracts deployments that want operational simplicity. Q3 growth driven by enterprise teams preferring managed runtime over self-hosted.
Managed-runtime growth04 — Enterprise DeploymentThe integration conversion.
The most actionable forecast number is enterprise pilot-to-prod conversion via MCP-integrated stacks. Q2 2026 baseline: overall conversion was 31%, with MCP-integrated pilots converting at 38% versus 22% for non-MCP. Both grow through Q3, with MCP-integrated pulling further ahead.
Enterprise pilot-to-prod conversion · MCP-integrated vs all
Forecast: weekly registry snapshots · client engagement data · Apr 202605 — Standards-War RiskThe fragmentation scenario.
The standards-war risk is the dominant downside scenario for the forecast. Anthropic remains the primary stewards of MCP, but Microsoft, OpenAI, and Google have shipped MCP-compatible servers and clients with proposed extensions in some cases diverging from core spec. Three failure modes worth tracking.
Extended-MCP fork
Major vendor ships incompatible extensionOpenAI or Microsoft formalizes 'extended-MCP' track that breaks compatibility with core spec. Servers split into two registries; clients have to support both. Worst-case for adoption; pulls Q3 forecast toward conservative scenario. Probability: 0.20.
Worst caseAuthentication standard fragmentation
Multiple competing auth proposalsMCP-bot-id, MCP-OAuth, MCP-mTLS, and proprietary alternatives compete for adoption. Enterprise security teams demand standardization; without it, procurement caution slows server adoption. Probability: 0.40.
Most likely fractureVendor pull-back from MCP
Major vendor de-prioritizes MCP commitmentGoogle, Microsoft, or OpenAI signals that MCP is one of several agent tool-use protocols rather than the default. Lowers vendor commitment to first-party server quality. Probability: 0.25.
Watch closely06 — What to WatchThe specific signals.
Five signals will resolve which scenario plays out. Watch them in this order through July and August.
July standards-track committee meetings
Public meeting notes from MCP working group sessions in July. If extension proposals resolve in core spec, base/aggressive scenario weight increases. If contested or referred to subgroup, conservative scenario weight increases.
Earliest signalEnterprise vendor first-party announcements
First-party MCP servers from HubSpot, Zendesk, Notion, ServiceNow, Workday, Slack. Q2 added 5 of these; Q3 should add 3–5 more. Vendor commitment is the leading indicator for curated-registry growth.
Vendor commitmentCloudflare AI MCP managed-runtime adoption
Managed-runtime growth signals enterprise preference for operational simplicity. If Cloudflare AI MCP crosses 1,000 by July, aggressive scenario weight increases. If it stays under 800, base scenario.
Operational signalAuthentication standard convergence
Watch for committee-approved authentication standard with 3+ implementations by August. Without it, enterprise security gates slow server adoption. With it, server registry growth accelerates.
Auth gateMajor-vendor pull-back signal
Watch for OpenAI, Microsoft, or Google de-prioritizing MCP first-party commitment in product announcements. Subtle signal but high consequence — pulls forecast strongly toward conservative scenario.
Tail risk07 — ConclusionThe base case is the operating assumption.
Plan against the base case; insure against the conservative scenario.
The base-case forecast (18,400 servers by Q3, 41% all-pilot conversion, 47% MCP-integrated conversion) is the most useful single number for integration roadmap planning. Enterprise teams should plan against MCP being the dominant tool-use plumbing through year-end and into 2027.
The conservative scenario insurance is harder to design but worth it. If a standards-war event hits in Q3, integration roadmaps that picked a single MCP server per integration look fine; integration roadmaps that hard-coded against a specific extension proposal look fragile. Pick first-party servers from curated registries where available; avoid betting on extensions that have not cleared committee.
We will publish the next forecast update in late July alongside the Q2 quarterly report. The Q3 actual count gets scored in early October, with a retrospective on the scenario that played out and recalibrated weights for the next forecast cycle.