Undermind is an AI co-researcher built for literature review and systematic scientific search. It targets academics, PhD students, and R&D teams who need to survey a large body of published research. Users describe an open-ended research goal, and Undermind returns a synthesized, citation-backed report on the relevant work.
The tool runs an iterative deep search. It queries semantic databases, evaluates hundreds of papers, and follows citation trails and key authors’ publication histories. It repeats this cycle until no new relevant papers appear, then reads full texts to pull details from figures, tables, and equations.
According to Undermind’s own benchmark, its deep search reaches 85 percent recall of the top 20 relevant papers within 10 minutes, versus 50 percent for GPT-5.6 Sol and 47 percent for Claude Opus 5 run as web search agents. Undermind also supports the Model Context Protocol, connecting to tools like Claude, ChatGPT, Cursor, VS Code, GitHub Copilot, Codex, and Grok.
Undermind was founded by two physics PhDs from MIT and is backed by Y Combinator. It reports use by scientists at MIT, Harvard, Caltech, Princeton, Berkeley, Cambridge, NIH, Novartis, and Toyota Research Center, plus more than 1,000 scientists at GSK.
Pricing
Undermind offers a Free plan with standard rate limits on chats and searches, and a Pro plan at $16 per month billed annually with 10x higher usage limits and deeper full-text analysis. A Team plan costs $15 per person per month billed annually and requires a minimum of 5 seats, adding centralized billing and priority support. Enterprise pricing is custom and adds SSO, an admin dashboard, increased compute, and dedicated support. Exact usage caps for the Free plan are not publicly disclosed.
* Disclaimer: Please note that pricing information may not be up to date. For the most accurate and current pricing details, refer to the official website.
Key Features
-
✓
Runs iterative deep searches across citation trails automatically
-
✓
Reads full-text PDFs to extract figures and equations
-
✓
Generates cited synthesis reports from evaluated papers
-
✓
Connects to outside agents via Model Context Protocol
-
✓
Finds a specific author’s entire body of work
-
✓
Sends alerts when new relevant papers are published
Use Cases
Comprehensive Literature Review
An academic explores an open-ended topic before starting a project. Undermind autonomously searches databases, author histories, and citation trails, then compiles a ranked list of the most relevant papers.
Validating Pilot Data
A researcher checks internal pilot data against published studies. Undermind reads full texts, isolates matching methodologies or statistics, and explains whether existing evidence supports or contradicts the findings.
Automated Evidence Synthesis
A medical professional builds an evidence review for a clinical treatment. Undermind processes multiple clinical trials and produces a table of measures and effects alongside a cited report.
Research Inside the IDE
A bioinformatician working in VS Code needs equations from recent papers. Through MCP, they call Undermind directly in the editor to search, read PDFs, and extract the equations.
Enterprise R&D Monitoring
A pharmaceutical team tracks a specific drug delivery mechanism. Undermind sends automated alerts whenever new relevant research is published, keeping the team current on the field.
Strengths & Weaknesses
Strengths
Undermind’s own benchmark reports 85 percent recall on key papers, ahead of GPT-5.6 and Claude Opus 5 running as web search agents.
It connects to widely used AI tools and coding environments, including Claude, ChatGPT, Cursor, and VS Code, via MCP.
Generated reports include inline citations that let researchers verify each statement against the source paper.
The company states that proprietary research data is never used to train AI models.
Undermind is used by more than 1,000 scientists at GSK and reports use by researchers at MIT, Harvard, and other institutions.
Weaknesses
The Free plan applies standard rate limits to chats and searches, with exact caps not publicly listed.
The deepest full-text analysis and most capable AI models require the paid Pro plan.
The Team plan requires a minimum of 5 seats, which totals at least $75 per month.
Programmatic batch processing through the API is limited to Enterprise customers.
Who Is This For?
Academic Researchers and PhD Students: useful for exhaustive literature reviews and confirming that no relevant prior work has been missed.
Pharmaceutical and Biotech R&D Teams: helps evaluate clinical trial data and track specific drug delivery research at scale.
Hard-Tech Entrepreneurs: helps founders assess technical landscapes and check an invention’s novelty against existing literature.
AI Developers and Enterprise IT: can connect Undermind through MCP or the Enterprise API to add a research layer to internal tools.
Frequently Asked Questions
How much does Undermind cost?
Undermind has a free plan, a Pro plan at $16 per month billed annually, a Team plan at $15 per person per month with a 5-seat minimum, and custom Enterprise pricing.
Is there a free plan?
Yes. The Free plan includes chat, deep search, report generation, workspace collaboration, and MCP agent connections, subject to standard rate limits.
What scientific databases does Undermind search?
Undermind has not published the specific list of databases, but describes semantic search across more than 200 million scientific papers.
Why does a deep search take several minutes?
Undermind runs an iterative multi-pass search, evaluating hundreds of papers, citation trails, and author histories until no new relevant results appear.
Does Undermind read full papers or just abstracts?
It reads full-text PDFs in parallel to extract data from figures, tables, and equations, in addition to searching abstracts.
Can I connect Undermind to tools I already use?
Yes. It supports the Model Context Protocol and connects to Claude, ChatGPT, Cursor, VS Code, GitHub Copilot, Codex, Grok, Antigravity, and Perplexity.
What is the difference between the Pro and Team plans?
Pro is for individual use and adds higher usage limits and deeper analysis. Team adds member management, priority support, and centralized billing for at least 5 seats.
Does Undermind support enterprise data sources?
Enterprise customers can commission a custom AI research platform that unifies internal documents with external literature.
Is API access available?
Programmatic API access, including batch processing, is offered as part of the Enterprise plan only.
Undermind connects through the Model Context Protocol and dedicated apps to ChatGPT, Claude, Cursor, VS Code, GitHub Copilot, Codex, Claude Code, Opencode, Grok, Antigravity, and Perplexity. Cursor’s web-based Cloud Agents are not yet supported. Any other MCP-compatible client can also connect using Undermind’s MCP server address.