Install in OpenAI Codex

Based on the Codex documentation as checked on 2026-11-09. Codex skills build on the same open Agent Skills format.

Skills

skills install @<namespace>/<slug> --client codex              # user scope
skills install @<namespace>/<slug> --client codex --project    # repository scope

By hand:

V=1.2.0; S=<slug>; N=<namespace>
curl -fsSLo $S-$V.tar.gz https://skills.spboucher.ai/api/v1/artifacts/$N/$S/versions/$V/download
curl -fsS https://skills.spboucher.ai/api/v1/artifacts/$N/$S/versions/$V/checksum | shasum -a 256 -c -
mkdir -p ~/.agents/skills/$S && tar -xzf $S-$V.tar.gz -C ~/.agents/skills/$S

We could not confirm whether the older ~/.codex/skills location is still read, so the registry does not install there.

A skill written for Claude Code may use frontmatter fields that are not part of the open specification. The release page lists them as a finding (skill-nonstandard-fields); Codex may ignore them.

MCP servers

Codex reads MCP servers from ~/.codex/config.toml (or .codex/config.toml in a trusted project). The artifact page generates the block for you.

Local server (stdio):

[mcp_servers.<name>]
command = "<command>"
args = ["…"]
env_vars = ["API_KEY"]     # forwarded from your shell environment

Remote server (Streamable HTTP):

[mcp_servers.<name>]
url = "https://example.org/mcp"
bearer_token_env_var = "THEIR_TOKEN"   # optional: name of an env var holding the token

For a stdio server you can also use codex mcp add <name> --env VAR=VALUE -- <command>. We show the TOML block for remote servers because we could not confirm the CLI flags for them.

The registry's own MCP server

[mcp_servers.skills-registry]
url = "https://skills.spboucher.ai/mcp"

Add bearer_token_env_var = "SKILLS_TOKEN" only if you want the publishing tools. See MCP reference.