med-paper-assistant: Skill for Claude Code

PICO Clinical Question Search

Use this workflow for clinical comparison questions. The MCP server does not semantically parse natural-language clinical questions. The agent extracts P/I/C/O, submits the structured handoff through parse_pico, and then executes the returned template: pico pipeline or an expanded Boolean query.

PICO Elements

Element	Meaning	Example
P	Population / patient group	ICU patients requiring sedation
I	Intervention / exposure / index test	remimazolam
C	Comparator, optional	propofol
O	Outcome, recommended	delirium, hypotension, sedation efficacy

Workflow

Agent extracts P/I/C/O
-> parse_pico(description, p, i, c, o)
-> optional generate_search_queries for P/I/C/O expansion
-> unified_search(query=original_question, pipeline=parse_pico.pipeline)

Step 1: Agent Extracts PICO

Do the semantic work in the agent. Do not ask the MCP server to infer P/I/C/O from free text. If P or I is unclear, ask the user before searching. Do not invent missing clinical details.

Step 2: Validate The Handoff

pico = parse_pico(
    description="Is remimazolam better than propofol for ICU sedation?",
    p="ICU patients requiring sedation",
    i="remimazolam",
    c="propofol",
    o="delirium, hypotension, sedation efficacy",
    question_type="therapy",
    sources="pubmed,europe_pmc",
    limit=50,
)

Expected useful fields:

• validation: whether required P/I fields are present
• pico: human-readable P/I/C/O
• query_elements: search fragments that the backend will use
• pipeline: ready-to-run template: pico YAML
• next_tool_call: suggested unified_search call

When only description is provided, parse_pico returns a schema and asks the agent to call it again with structured elements.