Runtime Adapters
One skill definition deployed to every AI runtime via MCP.
Runtime Adapters are the interoperability layer that makes Headways runtime-neutral. They compile a single skill definition into the native format of each AI runtime.
Supported Runtimes
| Runtime | Adapter Output | Status |
|---|---|---|
| Claude Code | CLAUDE.md context + MCP tool install | Active |
| GitHub Copilot | SKILL.md format | Coming soon |
| Cursor | .cursor/rules/ files | Coming soon |
| OpenAI Codex | Agent Skills specification | Coming soon |
| Custom (MCP) | Raw MCP tool definitions | Active |
How Adapters Work
- You define a skill once in Headways' universal format
- The Runtime Adapter compiles it to each target runtime's native specification
- When users sync skills, they get the version compiled for their runtime
- Governance rules travel with the skill, regardless of runtime
MCP as the Interop Layer
Headways uses the Model Context Protocol (MCP) as its primary interoperability mechanism. MCP is now a Linux Foundation standard with support from Anthropic, Google, Microsoft, and AWS.
Benefits of MCP-native design:
- Standard protocol means any MCP-compatible runtime works automatically
- No proprietary hooks or vendor-specific integrations
- Governance can be enforced at the protocol level
No Vendor Lock-in
This is Headways' primary defense against vendor up-stack movement. Each vendor governs their own runtime. Because Headways operates at the adapter layer, switching runtimes (or adding new ones) requires zero reconfiguration of skills or policies.