The MCP Ecosystem at 18 Months
Anthropic announced the Model Context Protocol (MCP) in November 2024 as an open standard for connecting AI assistants to data sources and tools. By Q2 2026, MCP has grown into one of the fastest-adopted open AI standards in history, with major client adoption from Anthropic, OpenAI, and developer-tool vendors. This page consolidates ecosystem statistics through Q2 2026.
Key Findings
- The MCP server directory and community-listed servers totalled approximately 8,000-12,000 distinct servers in Q2 2026, up from approximately 50 at launch in November 2024.
- Major MCP clients in production include Claude Desktop, Claude Code, Cursor, Windsurf, Zed, Continue, plus OpenAI's Custom GPTs (which added MCP support in 2025).
- The most-installed MCP servers connect to Slack, GitHub, Notion, Google Drive, Postgres, file systems, and web fetching, the standard developer toolset.
- Enterprise adoption of MCP grew faster than community adoption in late 2025; large enterprises increasingly run private MCP servers exposing internal data sources to AI assistants.
- MCP has effectively become the de facto open standard for AI-tool integration, displacing competing approaches (OpenAI Functions, custom plug-in standards) for cross-vendor interoperability.
MCP Server Directory Growth
| Period | Listed servers | Notable additions |
|---|---|---|
| November 2024 | ~50 | Anthropic launches reference servers (filesystem, GitHub, Slack, Postgres) |
| Q1 2025 | ~400 | Community contributions ramp; first vendor-built servers (Notion, Linear) |
| Q3 2025 | ~2,500 | OpenAI announces MCP support; ecosystem inflection |
| Q1 2026 | ~7,000 | Enterprise MCP servers proliferate; major SaaS vendors ship official servers |
| Q2 2026 | ~8,000-12,000 | Continued growth; signal-to-noise ratio degrades as low-quality servers proliferate |
Top MCP Servers by Reported Installation
| Server | Maintained by | Primary use |
|---|---|---|
| filesystem | Anthropic reference | Local file access |
| github | GitHub official | Repo, issue, PR access |
| slack | Slack official | Channel and message access |
| postgres | Anthropic reference | Database queries |
| notion | Notion official | Workspace access |
| google-drive | Google official | Drive and Workspace |
| fetch / puppeteer | Community | Web fetching |
| linear | Linear official | Issue tracker |
| memory / mem0 | Mem0 | Persistent memory |
| sentry | Sentry official | Error monitoring data |
MCP Client Adoption
| Client | MCP support added | Estimated user base |
|---|---|---|
| Claude Desktop | November 2024 (launch) | Multi-million |
| Claude Code | 2025 | Hundreds of thousands |
| Cursor | Q1 2025 | Multi-million |
| Windsurf | Q1 2025 | Hundreds of thousands |
| Zed | Q1 2025 | Smaller |
| Continue | Q2 2025 | Smaller |
| OpenAI Custom GPTs / API | Q3 2025 | Hundreds of millions (via ChatGPT) |
| Google Gemini Code Assist | Q4 2025 (announced) | Smaller |
| Microsoft Copilot Studio | 2026 | Enterprise-wide |
Enterprise MCP Adoption Patterns
Three patterns dominate enterprise MCP deployments:
- Internal-data MCP servers: enterprises ship private MCP servers exposing internal data lakes, knowledge bases, ticketing systems to employee AI assistants. Most common pattern.
- Vendor-published MCP servers: SaaS vendors (Snowflake, Databricks, Salesforce, ServiceNow, Atlassian, GitHub) publish official MCP servers for their platforms.
- MCP gateway / proxy patterns: enterprises operate MCP gateways with authentication, audit logging, and policy enforcement between AI clients and source MCP servers.
Quality and Discovery Challenges
The ecosystem's rapid growth has created real signal-to-noise problems:
- The official MCP directory plus community-curated lists differ in inclusion criteria; users install servers from various sources.
- Low-quality, broken, and abandoned servers proliferate; users report 30-50 percent installation failure rates on community servers.
- Security review is uneven; MCP servers run with substantial privileges and represent a meaningful attack surface that few enterprises audit systematically.
- Server documentation quality varies wildly; AI assistants struggle to use under-documented servers effectively.
Standards and Governance
MCP governance lives at modelcontextprotocol.io, with the specification maintained on the MCP GitHub organisation. Anthropic stewards the protocol but multi-vendor adoption has created de facto multi-stakeholder governance. Major spec updates in 2025-2026 added: structured tool annotations, streaming responses, OAuth-flow standardisation, and resource-quota negotiation.
Brand Visibility Implications
MCP servers are an emerging AI brand-visibility surface. When an MCP-enabled assistant queries data on the user's behalf, the brand whose MCP server connects becomes the canonical source. Vendors with official, well-documented, widely-adopted MCP servers gain disproportionate AI-mediated visibility because the assistant uses their data structure and naming. Brands without MCP servers are increasingly invisible inside MCP-enabled buyer journeys. See our MCP Brand Visibility FAQ for the operational implications.
Methodology
Server count from official MCP directory plus community-curated lists (mcp.so, awesome-mcp-servers GitHub, Smithery, Glama). Client adoption from public client documentation. Installation figures from public client telemetry where disclosed and Presenc AI deployment instrumentation. Enterprise patterns from observed deployments across 60+ enterprise customers. Updated quarterly.
How Presenc AI Helps
Presenc AI tracks brand-mention rates inside MCP-enabled AI assistant queries, distinguishing brand exposure that flows through MCP server data from exposure through general training data. For brands evaluating MCP server investment or measuring MCP-mediated brand visibility, this is the operational signal that connects MCP ecosystem participation to real brand-discovery outcomes.