MCP Adoption by Client: Claude Code Leads at 95%, Hugging Face Chat UI Hits 100%
Live from AGNTCon + MCPCon Europe Day 2 — new data shows how different MCP clients are adopting the protocol in production. Spoiler: the ecosystem is moving fast, and it’s not just the big AI companies leading adoption.
The Data: Protocol Adoption by Client
The slide, presented during Day 2 sessions, tracks MCP tool call adoption from July 28, 2026 through September 15, 2026 (the dates of AGNTCon + MCPCon Europe itself). The “Last 7” column shows adoption rates for September 9–15.
| Client | Last 7 (Sep 9-15) | Trajectory |
|---|---|---|
| Claude Code | 95.3% | Steady climb from 0% to 95% |
| Anthropic/ClaudeAI | 99.5% | Spike in mid-August to 100% |
| Hugging Face Chat UI | 100.0% | Spike in mid-August to 100% |
| openai-mcp | 32.6% | Volatile, peaked in August then dropped |
| openai-mcp (Codex) | 15.3% | Consistently low, jagged trajectory |
| Codex CLI | 0.0% | No adoption |
The purple bars at the bottom of each chart show daily volume — the intensity of the color indicates how many tool calls were routed through MCP on any given day.
What This Means
1. Claude Code is the MCP Gateway for Developers
At 95.3% adoption, Claude Code (Anthropic’s agentic coding tool) is the fastest-adopting MCP client among developer tools. This makes sense — Claude Code’s entire value proposition is tool use, and MCP is the protocol that connects Claude to your local tools, data, and APIs.
The steady climb from 0% in late July to 95% in mid-September shows sustained, organic adoption — developers discovering MCP through daily use of Claude Code.
2. Hugging Face Hits 100% — The Open Source Signal
Hugging Face Chat UI reached 100% MCP adoption — meaning every tool call through their interface routes through MCP. This is significant because Hugging Face represents the open source community’s primary gateway to AI agents.
The mid-August spike coincides with Hugging Face’s MCP server release and the broader ecosystem momentum from AGNTCon pre-events. 100% adoption from an open source platform is a strong signal that the community has standardized on MCP.
3. OpenAI’s Position: Mixed
The OpenAI ecosystem tells two stories:
- openai-mcp (the community MCP bridge): peaked at ~70-80% in August, then declined to 32.6% — suggesting teams may be moving to native MCP integration rather than bridge servers
- openai-mcp (Codex): stuck at 15.3% — Codex appears to have limited MCP integration
- Codex CLI: 0% adoption — no MCP integration at all
This is a notable gap. While Claude Code, Hugging Face, and Anthropic are all-in on MCP, OpenAI’s tools have limited or no MCP support. For teams standardizing on MCP, this creates a clear incentive to use Claude Code or Hugging Face over OpenAI’s toolchain.
Context: MCP in Production
This data comes at a critical inflection point. As I covered in my Day 1 recap, MCP graduated to enterprise infrastructure with a stateless core — the update that Shaun Smith (Hugging Face) demonstrated in his Day 2 keynote for production at scale.
With 89% of agentic AI teams using MCP (per the AAIF Member Pulse Survey), the question has shifted from “will MCP be adopted?” to “how fast can it scale?” The adoption-by-client data suggests:
- Developer tools (Claude Code) drive grassroots MCP adoption
- Open source platforms (Hugging Face) achieve full ecosystem integration
- Proprietary platforms that don’t natively support MCP risk falling behind
The Broader Stack
MCP doesn’t exist in isolation. Alongside it, the open agentic stack includes:
- A2A (Agent2Agent) — now joining the AAIF for agent-to-agent communication
- Skills — 69% adoption for task orchestration
- AGENTS.md — 51% adoption for agent instructions
- AGNTCY — 37% adoption for full lifecycle management
