research-primitives

name	research-primitives
description	Reference guide for core Research Harness public MCP operations
public_suite	research
task_type	reference
data_access_level	local
inputs	primitive name, workflow question, tool-surface question
outputs	public MCP guidance, safe primitive usage pattern
gates	prefer public MCP tools over CLI for research primitives
artifact_schemas	none
estimated_runtime	1-5 min
allowed-tools	["Read"]

Research Primitives

Reference guide for the core public Research Harness operations available via the contracted MCP surface. Additional registered primitives may exist as internal Python building blocks or compatibility-dispatch names.

---

paper_search

Attribute	Value
Category	RETRIEVAL
LLM Required	No
Idempotent	Yes

Description: Search for papers by query across configured providers (arXiv, Semantic Scholar, etc.)

Input:

• query (string, required): Search query
• topic_id (integer): Associate with topic
• max_results (integer): Default 20
• year_from/year_to (integer): Year range filter
• venue_filter (string): Venue substring filter

Output: PaperSearchOutput

• papers: List of PaperRef objects
• provider: Which provider served the query
• query_used: Actual query sent

Example:

mcp__research-harness__paper_search with query="transformer attention mechanism"

---

paper_ingest

Attribute	Value
Category	RETRIEVAL
LLM Required	No
Idempotent	Yes

Description: Ingest a paper into the pool by arxiv_id, doi, or pdf_path

Input:

• source (string, required): arxiv_id, DOI, or file path
• topic_id (integer): Associate with topic
• relevance (string): "high", "medium", or "low"

Output: PaperIngestOutput

• paper_id: Assigned ID
• title: Paper title
• status: "created" or "merged"
• merged_fields: Fields updated if merged

Example:

mcp__research-harness__paper_ingest with source="arxiv:2401.12345" relevance="high"

---

paper_summarize

Attribute	Value
Category	COMPREHENSION
LLM Required	Yes

Description: Generate a focused summary of a paper

Input:

• paper_id (integer, required): Paper to summarize
• focus (string): Specific aspect to focus on

Output: SummaryOutput

• paper_id: Source paper
• summary: Generated summary text
• focus: Focus area used
• confidence: Quality score
• model_used: Which model generated it

Example:

mcp__research-harness__paper_summarize with paper_id=123 focus="methodology"

---

claim_extract

Attribute	Value
Category	EXTRACTION
LLM Required	Yes

Description: Extract research claims from papers within a topic

Input:

• paper_ids (list[int], required): Papers to analyze
• topic_id (integer, required): Topic context
• focus (string): Optional focus area

Output: ClaimExtractOutput

• claims: List of Claim objects
• papers_processed: Count processed

Example:

mcp__research-harness__claim_extract with paper_ids=[1,2,3] topic_id=1

---

evidence_link

Attribute	Value
Category	EXTRACTION
LLM Required	No
Idempotent	Yes

Description: Link a claim to supporting evidence

Input:

• claim_id (string, required): Claim to link
• source_type (string, required): "paper" or "external"
• source_id (string, required): Source identifier
• strength (string): "strong", "moderate", or "weak"
• notes (string): Explanation

Output: EvidenceLinkOutput

• link: Created EvidenceLink
• created: Whether new link was created

Example:

mcp__research-harness__evidence_link with claim_id="claim_abc123" source_type="paper" source_id="1"

---

gap_detect

Attribute	Value
Category	ANALYSIS
LLM Required	Yes

Description: Detect research gaps in a topic's literature

Input:

• topic_id (integer, required): Topic to analyze
• focus (string): Optional focus area

Output: GapDetectOutput

• gaps: List of Gap objects
• papers_analyzed: Count analyzed

Example:

mcp__research-harness__gap_detect with topic_id=1 focus="evaluation methods"

---

baseline_identify

Attribute	Value
Category	EXTRACTION
LLM Required	Yes

Description: Identify baseline methods for comparison in a topic

Input:

• topic_id (integer, required): Topic to analyze
• focus (string): Optional focus area

Output: BaselineIdentifyOutput

• baselines: List of Baseline objects

Example:

mcp__research-harness__baseline_identify with topic_id=1

---

section_draft

Attribute	Value
Category	GENERATION
LLM Required	Yes

Description: Draft a paper section using linked evidence

Input:

• section (string, required): Section name/type
• topic_id (integer, required): Topic context
• evidence_ids (list[string]): Claims/evidence to include
• outline (string): Section outline
• max_words (integer): Target length (default 2000)

Output: SectionDraftOutput

• draft: DraftText object with content

Example:

mcp__research-harness__section_draft with section="related_work" topic_id=1

---

consistency_check

Attribute	Value
Category	VERIFICATION
LLM Required	Yes

Description: Check consistency across drafted sections

Input:

• topic_id (integer, required): Topic to check
• sections (list[string]): Specific sections to check (empty = all)

Output: ConsistencyCheckOutput

• issues: List of ConsistencyIssue objects
• sections_checked: List of sections analyzed

Example:

mcp__research-harness__consistency_check with topic_id=1 sections=["method", "results"]