Mcporter Skillifier

Use this skill to generate a Codex skill that calls an MCP server through the mcporter CLI. The generated skill contains a shell wrapper and a skill-local mcporter.json; it does not register MCP tools with Codex. The wrapper invokes bash scripts/mcp through exec_command.

The wrapper uses npx -y mcporter@${MCPORTER_VERSION} by default, with the generated script pin as its default version. Set MCPORTER_BIN=mcporter to use a local binary instead. MCPorter provides the keep-alive daemon used by servers that retain state between calls.

Generate the skill

Run scripts/generate_skill.py from this skill's directory. Choose exactly one connection method: --stdio for a local process or --http-url for an HTTP MCP endpoint.

For a stdio server that keeps an in-memory continuation, enable --keep-alive:

python3 scripts/generate_skill.py \
  --skill-name my-mcp-skill \
  --stdio 'uvx --from ${PAL_MCP_FROM} pal-mcp-server' \
  --keep-alive \
  --with-fixture-tests

The generator normalizes the skill and server names to hyphen-case. By default it writes beside this skill's directory; set --out-dir to choose another parent directory. The output directory name is the normalized skill name. Use --force only when you intend to replace an existing output directory.

Set server environment values with repeatable --env KEY=VALUE arguments. Optional --server-name and --server-description configure the entry in mcporter.json. For keep-alive servers, --idle-timeout-ms sets the daemon idle timeout.

--with-fixture-tests adds a deterministic fixture MCP server and offline tests for multi-turn behavior. Generation does not run the generated self-test unless --verify is supplied. Verification requires npx and may require authentication or network access; --no-verify explicitly keeps generation offline.

MCPorter interpolates ${VAR} in configuration commands, but does not support Bash defaults such as ${VAR:-fallback}. Set variables before running the generated skill when a value needs a default. Avoid ${local_var} in shell snippets embedded in mcporter.json; MCPorter may treat it as a required configuration placeholder.

Call the generated skill

From the generated skill directory, list the configured server's tools and call one:

bash scripts/mcp list
bash scripts/mcp call <tool> --output json --args '{"key":"value"}'

The wrapper also accepts fully qualified server.tool selectors and HTTP URLs after call. It uses the skill's mcporter.json by default; set MCP_SKILL_CONFIG to use another config file. With no command or a help argument, scripts/mcp prints its usage and the MCPorter selection options.

For in-memory continuation_id flows, generate with --keep-alive so separate wrapper invocations reuse the server process. The fixture tests exercise this behavior without API keys:

python3 -m unittest -q tests/test_offline.py

The codex-skill-scoped-mcp experiment also contains integration checks for continuity across CLI invocations, isolation between skill configs, parallel calls, and parallel invocations:

• experiments/codex-skill-scoped-mcp/scripts/test_mcporter_skillifier_fixture_multiturn.sh
• experiments/codex-skill-scoped-mcp/scripts/test_mcporter_skillifier_parallel_calls_same_daemon.sh
• experiments/codex-skill-scoped-mcp/scripts/test_mcporter_skillifier_parallel_invocations_isolated_daemons.sh