Skip to main content
An adapter connects facets to one AI coding tool. It knows where that tool keeps each kind of asset: files for skills, agents, and commands, and configuration entries for MCP servers. Install one adapter per tool you use, then every facet you add materializes into all of them. Adapters are installed per machine, not per project:
See facet adapter add for sources, versions, and compatibility recovery.

​
Where assets land

Text assets are written at project scope, inside your checkout, so they travel with the repository. MCP servers land in configuration instead, covered below.
​
first-party
.claude/skills/<name>/SKILL.md, .claude/agents/<name>.md, .claude/commands/<name>.md
​
first-party
.opencode/skills/<name>/SKILL.md, .opencode/agents/<name>.md, .opencode/commands/<name>.md
​
first-party
.agents/skills/<name>/SKILL.md, .agents/commands/<name>.md, and .codex/agents/<name>.toml
The name on disk is the effective name, which is the name your project chose if you aliased the text asset.

​
Where MCP configuration lands

​
first-party
.mcp.json
​
first-party
Whichever of .opencode/opencode.jsonc, .opencode/opencode.json, opencode.jsonc, or opencode.json already holds your servers
​
first-party
.codex/config.toml
Your unrelated settings, comments, and any server entry the project does not own are preserved. See MCP servers.

​
Building your own

An adapter is a small npm package that plans file changes and returns them; the CLI performs every write. See the Custom adapters guide to build one, and the Adapter SDK reference for the contract.

​
Further reading