Noodle MCP: Biomedical Literature Discovery for Agents

Where this fits

You're building an agent that helps a researcher verify a claim about a gene-disease association. The user says, "Find papers that support this and show me how they cite each other." Without a dedicated tool, your agent would scrape PubMed HTML, guess at IDs, and miss the citation graph. Noodle Biomedical Literature Discovery MCP solves this by giving your agent a hosted, read-only API for searching, resolving PMIDs/PMCIDs/DOIs, and traversing citation or semantic links—no Noodle setup required.

Why agents benefit

• Natural-language search: Route ambiguous queries like "recent papers on CRISPR off-target effects" to search_biomedical_literature and get structured results with provenance.
• Exact identifier resolution: Convert between PMID, PMCID, and DOI instantly, so your agent can link a user's reference to full work records.
• Graph traversal: Walk citation or semantic neighborhoods from a resolved PMID, preserving edge types and provenance—useful for literature reviews or 'related work' generation.
• Corpus metadata: Answer freshness and coverage questions (e.g., "how many papers from 2024 are indexed?") via get_corpus_summary.
• Safety guardrails: The skill enforces public-only queries and warns against inferring causality from proximity, keeping your agent honest.

Practical scenarios

Scenario 1: Verify a claim with source-linked evidence

A user asks, "Is there evidence that metformin reduces cancer risk?" Your agent calls search_biomedical_literature, gets a list of papers with snippets and links, then uses inspect_work to pull abstracts. It can even trace citations to see which papers reference the key study, building a mini evidence chain.

Scenario 2: Resolve a messy citation

The user pastes a DOI that's actually a PMCID. Your agent uses the identifier lookup tool to normalize it, then fetches the full record and related papers—no regex gymnastics.

Scenario 3: Build a literature map for a research brief

Your agent starts from a known PMID, uses get_related_papers to expand outward, and returns a graph of connected works with edge types (citation, semantic similarity). It keeps a visited set to avoid loops and stops when neighborhoods are missing—perfect for a scoping report.

Add it to your agent workflow

Install the skill via OpenAgentSkill CLI:

npx skills add helena-bioinformatics/noodle-mcp --skill noodle-biomedical-literature-discovery-mcp

Then in your agent's tool configuration, expose the MCP tools. A simple example call:

{
  "tool": "search_biomedical_literature",
  "input": {
    "query": "gut microbiome and Parkinson's disease",
    "limit": 10
  }
}

The response includes paper IDs, titles, and source links—ready to feed into your agent's reasoning loop.

Compare before adopting

Before committing, compare against alternatives: PubMed E-utilities (free but requires API key and no graph traversal), Semantic Scholar API (graph but no PMCID resolution), and other MCP servers. Evaluate quality signals like freshness of the corpus (how often is it updated?), the precision of identifier resolution, and the clarity of returned provenance. Also consider whether your workflow needs citation graph traversal or just simple search—Noodle shines when you need both.

Why it is worth tracking

This skill is early (0 stars, single maintainer), but it addresses a real gap: a hosted, read-only biomedical literature MCP with graph traversal and identifier normalization. Its Apache-2.0 license and clear documentation make it easy to evaluate. Watch for updates to the underlying Biorouter corpus and community adoption. If your agent handles biomedical research queries, this is worth a test run now—and a bookmark for when it matures. You can explore its full profile on OpenAgentSkill.