Jaypore Labs
Back to journal
Engineering

Beyond MCP: tool-use specs in major models

MCP isn't the only way. Each major model has its own tool-use spec. Knowing the landscape matters.

Yash ShahApril 9, 20262 min read

MCP is the standard, but each major model has its own tool-use spec — historically and ongoing. A team building cross-provider applications needs to know the landscape.

The state of the art

MCP. Anthropic-originated; widely adopted. The standard for ecosystem tools.

Function calling (OpenAI). Predates MCP; still widely used in OpenAI-specific work. Compatible patterns; different protocol.

Tool use (Anthropic API). Anthropic's native tool-use API. MCP-compatible at the spec level.

Native function/tool calling in other providers. Each has its own variant.

Reviewer ritual

For cross-provider applications:

  • Document the tool-use approach per provider.
  • Plan migration if standards converge.
  • Avoid lock-in to a specific provider's spec.

A real comparison

A team building tools for both Claude and ChatGPT:

  • Built MCP servers (the standard).
  • Wrapped MCP for ChatGPT via a translator (function-calling format).
  • Native MCP for Claude.

The team's investment in MCP servers paid off. The wrapping work was straightforward.

Trade-offs

  • Stick to MCP: forward-looking; some translation needed for non-MCP providers.
  • Per-provider native specs: less translation; more lock-in.

The industry is converging on MCP. Lean MCP.

Limits

MCP doesn't yet cover:

  • Some edge cases other specs handle differently.
  • Multi-modal arguments (images in tool calls).
  • Some streaming patterns.

These gaps are closing as MCP evolves.

What we won't ship

Cross-provider applications without an abstraction layer.

Lock-in to a specific provider's spec when MCP would do.

Skipping the migration plan when standards evolve.

Close

Beyond MCP, the landscape has multiple tool-use specs. MCP is the de-facto standard. Building MCP-first with provider-specific adapters where needed is the durable strategy.

Related reading


We build AI-enabled software and help businesses put AI to work. If you're navigating tool-use specs, we'd love to hear about it. Get in touch.

Tagged
MCPTool UseEngineeringStandardsComparison
Share