Skip to content

Adapter Reference

An adapter encapsulates the format of a specific AI environment.

Summary

AI environmentTarget IDDefault base dirSkillsMCP config
Claude Codeclaude-code~/.claudeskills/<name>/SKILL.md~/.claude.json
Cursorcursor~/.cursorskills/<name>/SKILL.mdmcp.json
Codex CLIcodex~/.codexskills/<name>/SKILL.mdconfig.toml

Claude Code

MCP servers are written to ~/.claude.json, in the mcpServers JSON section — this is the file Claude Code reads for MCP server definitions (not settings.json, which stores policies like allowedMcpServers):

json
{
  "mcpServers": {
    "context7": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-mcp"],
      "env": {
        "API_KEY": "..."
      }
    }
  }
}

Cursor

Cursor uses the mcpServers JSON format in mcp.json.

The entry structure is the same as for Claude Code.

Cursor may create ~/.cursor/mcp.json as an empty file before any server is configured. AIM treats an empty config file as an empty object {} — MCP server installation continues correctly.

Codex CLI

Codex CLI uses TOML:

toml
[mcp_servers.context7]
command = "npx"
args = ["-y", "@upstash/context7-mcp"]

[mcp_servers.context7.env]
API_KEY = "..."

Overriding paths

If an environment is installed in a non-standard location, set the base dir in aim.local.yaml:

yaml
environments:
  claude-code: /custom/path/.claude

The adapter itself knows where skills and the MCP config are located within the base dir.

Released under the Apache 2.0 License.