MCP Ecosystem

The Model Context Protocol ecosystem in 2026: governance, spec evolution, and enterprise adoption.

Governance

MCP was donated to the Agentic AI Foundation (AAIF), a Linux Foundation directed fund, in December 2025. Co-founded by Anthropic, Block, and OpenAI.

This means:

MCP is vendor-neutral, not Anthropic-controlled
Spec changes go through an open governance process
Multiple organizations contribute to the standard
Enterprise adoption is accelerating because of neutrality

Spec Evolution

Streamable HTTP Transport

The original remote transport was Server-Sent Events (SSE). The new standard is streamable HTTP:

	SSE (legacy)	Streamable HTTP (new)
Connection	Long-lived	Request-response with optional streaming
Scalability	Stateful sessions, hard to load balance	Stateless, standard HTTP infrastructure
Firewall-friendly	Often blocked	Standard HTTPS
Resumability	Manual reconnection	Built into protocol

SSE still works but streamable HTTP is the recommended transport for new remote servers.

Tasks Primitive

Async operations that may take a long time:

{
  "method": "tasks/create",
  "params": {
    "name": "process-large-dataset",
    "timeout": 3600,
    "retry": { "maxAttempts": 3, "backoffMs": 1000 }
  }
}

Tasks support:

Progress reporting
Cancellation
Retry semantics with exponential backoff
Expiry policies
Status polling

MCP Apps

Tools can now return interactive UI components:

{
  "type": "tool_result",
  "content": [
    {
      "type": "app",
      "component": "data-table",
      "props": {
        "columns": ["Name", "Status", "Updated"],
        "rows": [...]
      }
    }
  ]
}

The client renders the component natively. This bridges the gap between text-only tool outputs and rich interactive experiences.

Elicitation

Servers can request structured input from users mid-task:

{
  "method": "elicitation/request",
  "params": {
    "message": "Which environment should I deploy to?",
    "options": ["staging", "production"],
    "type": "select"
  }
}

Instead of the model guessing or asking through natural language, the server directly prompts the user through a structured dialog. Supported types: text, select, multi-select, confirm.

Enterprise Readiness

Audit Trails

Every tool call, resource access, and prompt use is logged with:

Timestamp
Caller identity
Input parameters
Output summary
Duration

SSO Integration

MCP servers can delegate authentication to enterprise SSO providers:

{
  "auth": {
    "type": "oauth2",
    "provider": "okta",
    "scopes": ["read:data", "write:data"]
  }
}

Gateway Behavior

MCP gateways sit between clients and servers, providing:

Rate limiting
Request/response logging
Tool-level access control
Usage metering

Config Portability

Standardized configuration format for MCP server connections:

{
  "mcpServers": {
    "github": {
      "transport": "streamable-http",
      "url": "https://mcp.github.com/v1",
      "auth": { "type": "bearer", "token": "${GITHUB_TOKEN}" }
    }
  }
}

Same config works across Claude Code, Cursor, and other MCP-compatible clients.

2026 Roadmap

Priority	Status	Description
Transport scalability	Shipped	Streamable HTTP replacing SSE
Tasks primitive	Shipped	Async operations with retry/expiry
MCP Apps	In progress	Interactive UI components
Elicitation	In progress	Server-initiated structured input
Governance maturation	Ongoing	AAIF working groups, RFC process
Enterprise readiness	Ongoing	Audit, SSO, gateways

Claude-Specific MCP Integration

Claude Code has deep MCP integration as a client:

Auto-discovers MCP servers from .claude/settings.json
Supports stdio (local) and streamable HTTP (remote) transports
Wildcard permissions for MCP tools
Skills compose with MCP (skills = workflows, MCP = tools)

See also: FastMCP for building MCP servers in Python.

API deep dive - Claude API features that interact with MCP
Skill patterns - how skills orchestrate MCP tools