Back to Browse

Evidence MCP Server

Developer ToolsModerate6.6MCP RegistryLocal
Free

Server data from the Official MCP Registry

Power tools for AI agents to improve and speed up medical literature research

About

Power tools for AI agents to improve and speed up medical literature research

Security Report

6.6
Moderate6.6Moderate Risk

Valid MCP server (3 strong, 4 medium validity signals). 6 known CVEs in dependencies (0 critical, 2 high severity) Package registry verified. Imported from the Official MCP Registry.

5 files analyzed · 7 issues found

Security scores are indicators to help you make informed decisions, not guarantees. Always review permissions before connecting any MCP server.

Permissions Required

This plugin requests these system permissions. Most are normal for its category.

file_system

Check that this permission is expected for this type of plugin.

HTTP Network Access

Connects to external APIs or services over the internet.

How to Install

Add this to your MCP configuration file:

{
  "mcpServers": {
    "evidence-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "evidence-mcp@1.0.17"
      ]
    }
  }
}

Some servers take additional arguments. See the repository for details.

Documentation

View on GitHub

From the project's GitHub README.

Version License MIT Forever Healthy npm MCP Registry

Forever Healthy Evidence MCP

Forever Healthy Evidence MCP

Power tools for AI agents to improve and speed up medical literature research. Developed while building evipedia.ai and experiencing the pain points and shortcomings of existing solutions.

Toolset:

  • A self-building local library of sources: PMID, DOI, DailyMed, NCT
  • Decider-based claim verification against those sources
  • Optimized access to PubMed, ClinicalTrials.gov & DailyMed
  • Optimized web fetch.

All functions are token-optimized to minimize context use, return agent-ready content, and cache locally for fast responses. The MCP is self-explanatory to agents, so they can start using it right away with no coding required.

Released under the MIT license to empower, simplify, and speed up medical literature research for everyone.

Evidence MCP Tools

Self-Building, Local Source Library

  • Builds itself as agents work
  • Supports pmid, doi, dailymed, and nct IDs
  • Local caching of metadata, abstract, and full text
  • Token-saving partial reads: agents read the outline, then pick only the sections they need
  • Full text fetched automatically wherever it can be obtained
  • Parallel discovery cascade for full text (PMC, Europe PMC, TDM, Unpaywall, OpenAlex, CORE, institutional proxy)
  • Support for Elsevier and Wiley TDM
  • Automatic conversion (HTML, XML, PDF) to token-optimized Markdown

Decider-Based Claim Verification

  • Fast verification of claims against library sources using decider models
  • Automatic choice of abstract or full text, depending on availability
  • Local caching for repeat queries: a cached verdict in under 1 ms vs. ~1 s from the model
  • Support for Perplexity, Google, OpenAI, and Liquid AI decision models

Optimized PubMed Access

  • Up to 97% fewer tokens than other PubMed MCPs (Anthropic's, cyanheads'): a 20-row search ~50–97% smaller, 20 records ~96% smaller
  • Local caching: a repeated query in milliseconds instead of a second

Optimized ClinicalTrials.gov Access

  • Up to 98% fewer tokens than other ClinicalTrials.gov MCPs (Anthropic's, cyanheads'): a trial record ~97% smaller, a search ~40% smaller than Anthropic's
  • Local caching: a repeated query in milliseconds instead of a second

Optimized DailyMed Access

  • FDA drug labels found by generic or brand name, narrowed to one manufacturer if needed: set ID, manufacturer, date and version per label
  • 98% fewer tokens than DailyMed's search page
  • Local caching: a repeated query in milliseconds

Optimized Web Fetch

  • Significantly higher success rate than the standard fetch
  • 20% more relevant content per result than the standard fetch
  • Local caching
  • Automatic use of the download cascade

Versatile Download Cascade

  • Multi-stage download cascade for web pages and full texts
  • Significantly higher success rate than the standard fetch
  • Escalates stage by stage, with parallel requests in each: fetch, browsers, unlockers
  • Fetch: plain fetch and a fetch proxy
  • Browsers: Playwright browser servers and a real Chrome
  • Unlockers: ZenRows, Zyte, Bright Data
  • Remembers failing targets and services and skips them for a while
  • Agent-deployable: comes with instructions for agents to set up and test every stage

Local Caching

  • All results cached locally where sensible
  • Response times in milliseconds instead of seconds
  • Extremely helpful when iterating multiple times over the same document
  • Automatic cache and refresh management

Log Viewer

  • Python-based TUI for operations monitoring

Tools

Tool
fh_info()version, library counts by type and level, sources, browsers, decider, configuration, cache hits
fh_library_add(id)download and store one record — pmid:, doi: (paper), dailymed: (FDA drug label), nct: (trial record); a paper with its full text wherever it can be obtained
fh_library_status(id)the status of one record (level, citable, corrections)
fh_library_read(id, section?, query?, max_chars?, max_hits?)the outline, one section or the passages matching a query; adds a missing record first
fh_verify_claim(claim, id)does this source state this claim? A verdict from the decider, on full text or abstract; adds a missing record first
fh_pubmed_search_articles(query, max_results?, sort?, years?)search PubMed in its own syntax: PMID, title, first author, year, journal, publication types, PMCID, library level
fh_pubmed_fetch_articles(pmids)title, first author, year, publication types, retracted and level for up to 200 PMIDs at once
fh_clinicaltrials_search_studies(query | intervention | condition, status?, sort?, max_results?)search ClinicalTrials.gov: NCT ID, title, status, phase, enrollment, dates, primary outcome, conditions, interventions
fh_clinicaltrials_get_study_record(nct)one trial as a review needs it: status, design, enrollment, dates, outcomes, eligibility, result publications
fh_clinicaltrials_get_study_results(nct, max_chars?)a trial's posted results as Markdown, capped
fh_dailymed_search_labels(drug_name, labeler?, max_results?)find FDA drug labels by name: set ID, title, manufacturer, published date, version
fh_fetch_url(url, max_length?, start_index?, raw?, title_only?)a web page or PDF as Markdown (XML, JSON and other text as is), cached, through the download cascade; same answer as the standard fetch tool

Quick Start

  • Add the MCP server
{
  "mcpServers": {
    "evidence-mcp": {
      "command": "npx",
      "args": ["-y", "evidence-mcp@latest"]
    }
  }
}

A .env elsewhere than your project root: add "--env", "/path/to/.env" to args.

Works out of the box with the keyless sources (PubMed, PMC, Europe PMC, Unpaywall, OpenAlex, plain fetch); only claim verification needs a Perplexity, Google, OpenAI, or Liquid AI key, or a local Decider.

Configuration (optional)

  • Clone this repository
  • Put a copy of <clone>/.env.example as .env in your project root, or tell your agent to add its content to your existing .env

Decider Model (optional, for claim verification)

Download Cascade (optional)

  • Configure the desired download services (fetch proxy, Playwright servers, Chrome, unlockers)
  • Tell your agent to run <clone>/utilities/deploy.sh up from your project folder (or with --env <your .env>) to deploy the configured cascade
  • Tell your agent to test the cascade for functionality and to detect accessibility issues (firewall, etc.)

Monitoring

  • Run <clone>/tui/run.sh from the folder that holds your .env (requires Python 3)

Demo Skill

/ev is a Claude Code skill that shows the server at work on any document that cites medical sources. It ships with this repository: clone the repository and open Claude Code in your clone.

The samples folder holds a demo document, telmisartan.md (an evidence review of telmisartan), to try the commands on.

The server stores its library in the evidence/ folder of the clone.

Command
/ev or /ev helpthe commands
/ev infoserver, configuration and library contents
/ev ingest <file>every source the document cites (PubMed, DOI, trials, drug labels) added to the library → <file>_sources.md
/ev verify <file>every claim the document makes, checked against the source it cites → <file>_claims.md

The file name is enough: documents are looked up in samples/. With the sample:

/ev ingest telmisartan.md   → samples/telmisartan_sources.md
/ev verify telmisartan.md   → samples/telmisartan_claims.md

Put your own document in samples/ and use its filename.

Documentation

License

MIT — © 2026 Forever Healthy Foundation

Reviews

No reviews yet

Be the first to review this server!