Model Context Protocol
MCP is what turns "ChatGPT mentioned your business" into "ChatGPT booked an appointment with your business". Different verb, different stakes.
Battle-tested in production. Build on it with confidence.
Agent protocol·Open-source
What It Is
The Model Context Protocol (MCP) is an open specification, originally developed by Anthropic, for connecting AI assistants to external systems. An MCP server exposes tools, resources, and prompts that any compatible AI assistant (Claude, ChatGPT, others) can discover and invoke. Adoption accelerated through 2025 as major engines added native MCP support.
Why It Matters
For AEO buyers, MCP is the bridge between visibility and action. Citation puts your business in the answer. MCP puts your business in the workflow. A travel agency cited in Claude is an outcome. A travel agency that Claude can directly search availability and book against via MCP is a customer flow.
As agent-driven actions become more common, businesses with MCP-exposed services get a fundamentally different relationship with AI assistants than businesses limited to text mentions.
Key Developments
- 2025: Major AI engines (Claude, OpenAI, others) shipped native MCP client support.
- 2025: Reference MCP servers for common business systems (calendars, CRMs, e-commerce) became widely available.
- Nov 2024: Anthropic introduced and open-sourced the MCP specification.
What to Watch
Watch which competing protocols emerge to challenge or complement MCP. A2A, OAGP, and ACP each occupy adjacent territory. Track which AI engines support MCP natively vs require plugin layers. Watch the security and trust frameworks around MCP server publication. Agent actions on real systems require verification, attestation, and rate limiting that the spec is still maturing on.
Strengths
- Open standard: Vendor-neutral. One MCP server works across multiple AI assistants.
- Action-enabling: Goes beyond citation to actual customer transactions. Booking, querying, purchasing.
- Strong incumbent backing: Anthropic-developed, OpenAI-supported, broad ecosystem participation.
- Developer-friendly: Reference implementations and SDKs in major languages lower the bar for businesses to expose services.
Considerations
- Discoverability gap: No central directory of MCP servers. Assistants need to be configured to connect, which limits consumer reach.
- Security maturity: Authentication, authorisation, and rate-limiting patterns are still evolving. Production deployments require careful review.
- Schema-first: MCP describes capabilities via JSON schemas, which works for structured tools but is awkward for natural-language workflows.
- Competition: A2A, OAGP, and others target the same problem from different angles. The eventual winner is unsettled.
Model Context Protocol· ACP — Agent Communication Protocol· A2A Protocol· Computer Use· Agent discovery manifests· OAGP — Open Agent Gateway Protocol