MCP Prompts

The Hubify MCP server includes pre-built prompts that guide AI assistants through common research workflows. These prompts include structured instructions and automatically inject relevant lab context.

Available Prompts

analyze_experiment

Guides the assistant through a structured analysis of experiment results.

Arguments:

Name	Type	Required	Description
experiment_id	string	Yes	Experiment to analyze

What it does:

1. Reads experiment results and QC report
2. Compares results to prior experiments
3. Identifies significant findings and anomalies
4. Proposes follow-up experiments (5-15 per the Houston Method)
5. Suggests knowledge base updates

Example invocation:

Use the analyze_experiment prompt for EXP-054

draft_section

Guides the assistant through drafting a paper section.

Arguments:

Name	Type	Required	Description
paper_id	string	Yes	Target paper
section	string	Yes	Section name (e.g., results, discussion)

What it does:

1. Reads the paper outline and claims table
2. Fetches relevant experiment results
3. Searches the knowledge base for supporting context
4. Generates a section draft in LaTeX
5. Cross-references claims against evidence

design_experiment

Helps design a new experiment with proper configuration.

Arguments:

Name	Type	Required	Description
goal	string	Yes	What the experiment should investigate

What it does:

1. Reviews the lab's research history for related work
2. Suggests experimental parameters and methodology
3. Recommends GPU type and estimated runtime
4. Defines QC gate criteria
5. Generates an experiment.yaml config file

review_results

Performs a structured review of recent results.

Arguments:

Name	Type	Required	Description
since	string	No	Time period (default: 7d)

What it does:

1. Lists all experiments completed in the time period
2. Summarizes findings and their significance
3. Identifies patterns across experiments
4. Flags results that contradict expectations
5. Recommends priority adjustments to the task queue

knowledge_synthesis

Synthesizes knowledge base entries on a topic into a coherent summary.

Arguments:

Name	Type	Required	Description
topic	string	Yes	Topic to synthesize

What it does:

1. Searches the knowledge base for relevant entries
2. Combines entity, concept, source, and comparison entries
3. Produces a structured summary with citations
4. Identifies gaps in the knowledge base
5. Suggests new entries to create

Using Prompts

In Claude Code or other MCP-compatible assistants, you can invoke prompts directly:

Use the design_experiment prompt with goal "Test whether quintom-B dark energy model
fits DESI BAO data better than LCDM"

The assistant receives the prompt template with your lab's data injected, then follows the structured workflow to produce high-quality output.

Custom Prompts

You can add custom prompts to the MCP server:

hubify mcp prompt add \
  --name "weekly_summary" \
  --template "Summarize all experiments from the past week. Include: results, cost, QC outcomes, and recommended next steps." \
  --args "weeks:number:1"

Custom prompts are stored in your lab configuration and available to all MCP hosts.