pal
Use the `pal` MCP server via MCPorter (with optional keep-alive daemon for multi-turn tools).
Files5
View on GitHubSKILL.mdagent-marketplace/src/experimental/skills/pal/SKILL.md
GitHubPal MCP
Use this skill to list or call tools exposed by the pal MCP server through
the bundled MCPorter wrapper.
Call tools
List the server's tools:
bash "scripts/mcp" list
Call a tool by its short name. Pass arguments as JSON:
bash "scripts/mcp" call <tool> --output json --args '{"key":"value"}'
The wrapper prefixes short names with pal.. It also accepts a fully qualified
server.tool name or an HTTP URL when a specific target is needed. Run
bash "scripts/mcp" --help for wrapper usage.
Continue multi-turn calls
When a PAL tool returns a continuation_id, make the next call through this
wrapper. Its MCPorter configuration uses the daemon-managed keep-alive
lifecycle, which reuses the server process and its in-memory continuation state
between calls. This lifecycle is required for those multi-turn flows.
Configure the server
The wrapper uses npx to run the pinned MCPorter version by default. The
server configuration is in mcporter.json.
| Variable | Effect |
|---|---|
MCPORTER_BIN |
Use this local MCPorter executable from PATH instead of npx (for example, mcporter). |
PAL_UVX_BIN |
Use this uvx executable path to launch PAL. When unset, the config searches PATH and common install paths. |
PAL_MCP_FROM |
Override the source from which uvx fetches the PAL server. |
PAL_DEFAULT_MODEL |
Set PAL's DEFAULT_MODEL; defaults to auto. |
The default launch requires npx, Node, uvx, and network access to the
configured server source. PAL also needs the provider credentials required by
the selected model in its environment, such as OPENAI_API_KEY,
GEMINI_API_KEY, or OPENROUTER_API_KEY.
Check the connection
Run the self-test to list PAL tools and call its version tool:
bash "scripts/selftest"