MCP Server
Generate Codify configs using Claude, Codex, and other AI coding tools
Codify runs a remote MCP server that gives any MCP-compatible AI coding tool direct access to Codify's resource library. Connect it to Claude Code, Codex, Cursor, or another MCP client, and the tool can look up real resource schemas and generate valid codify.jsonc files.
Connecting
The server is available over Streamable HTTP at:
https://api.codifycli.com/v1/mcpNo API key or authentication is required.
Claude Code:
claude mcp add --transport http codify https://api.codifycli.com/v1/mcpCodex CLI: Codex has no CLI shortcut for HTTP servers yet, so add the server directly to ~/.codex/config.toml:
[mcp_servers.codify]
url = "https://api.codifycli.com/v1/mcp"Other tools: Check your tool's MCP documentation for how to add a remote HTTP server. Tools that only support stdio servers need a local bridge such as mcp-remote pointed at the URL above.
Tools
| Tool | Description |
|---|---|
search_resources | Semantically search resource types by capability (e.g. "install Node.js"). |
get_resource_schema | Get the full schema, parameters, and example configs for one resource type. |
list_resources | List all available resource types in the default plugin. |
search_templates | Semantically search pre-built config templates by use case. |
get_template | Fetch a full template config by name or ID. |
generate_config_hint | Given a description of what you want, return the resource schemas and examples needed to write it. |
The server also sends config authoring rules (valid file naming, when to use dependsOn and os gating, how to write idempotent action conditions, and so on) as MCP server instructions. Clients that support this pick it up automatically, so the AI tool is grounded in Codify's actual conventions rather than inventing its own.
Example
You ask your AI coding assistant:
"Set up a Node.js and Docker dev environment"
The assistant calls search_resources for "Node.js" and "Docker", finds nvm and docker, then calls get_resource_schema for each to confirm the correct parameters. It writes the result to codify.jsonc:
[
{
"type": "nvm",
"version": "22"
},
{
"type": "docker"
}
]Connecting the MCP server only helps the AI tool write the config. Always run codify plan to review the changes, then codify apply to confirm before anything is installed.