# Undermind — AI co-researcher that runs deep literature searches over 200M+ scientific papers and cites every source.

> Source: The Agents Index — https://theagentsindex.com/undermind (structured, researched, re-verified)
> Facts last verified: 2026-09-26

Undermind is an AI literature research agent that plans and runs multi-stage searches across a database of over 200 million scientific articles, reads full texts and figures, and writes cited reports. It runs in a web app and connects to other agents over MCP.

| Fact | Value |
| --- | --- |
| Website | https://www.undermind.ai |
| Price | from $20/month billed monthly, $16/month billed annually |
| Free tier | Yes |

Free plan at $0 for exploratory research: strong AI models for chat, deep searches and reports, shared workspaces, agent connection, and 'Standard rate limits on chats and searches'. Search covers abstracts and metadata only - full-text search of open access and preprint articles requires Pro.

## What it does

**Done-for-you service** · Partly — Mostly self-serve software, but the Enterprise page also offers a delivered custom research platform built by Undermind's own team within weeks.

> Fast deployment: Within weeks, our team delivers custom builds tailored to your unique needs and data sources
> — https://www.undermind.ai/enterprise · Undermind for organizations section, Custom AI research platform card

## How much it does unattended

**Runs autonomously** · Yes — The deep search plans its own query sequence, follows citations and authors and decides itself when to stop, with no human step between submission and result.

> Runs a deep literature review from an open-ended research goal and produces a ranked list of papers with a written synthesis. Plans its own searches, follows citations and key authors, and stops only when new searches stop finding relevant papers.
> — https://www.undermind.ai/mcp · MCP tool list, Find section, launch_deep_search entry

**Multi-agent** · Partly — Undermind is the research layer another agent orchestrates over MCP; it does not advertise several Undermind agents coordinating on one task.

> MCP integration: Undermind plugs in as a literature research layer for the agentic systems your teams are building
> — https://www.undermind.ai/enterprise · Undermind for organizations, 'Undermind API' bullet list

**Agent permissions** · Partly — Scope is agreed up front through follow-up questions rather than by approving individual actions; no per-action permission model is published.

> Undermind asks follow-up questions to understand exactly what you need.
> — https://www.undermind.ai/ · 'How Undermind works' section, step 1 — Describe

## Which models it runs on

**Claude** · Yes — Published Claude connector, one-click setup, plus documented Claude Code CLI install.

> Open Undermind in Claude’s connector directory. If the button does not open it, paste https://claude.ai/directory/connectors/undermind into your browser.
> — https://www.undermind.ai/mcp · Claude — Web & Desktop section, setup prose

**GPT** · Partly — Named as the reasoning engine only in the January 2024 whitepaper, which the homepage itself labels the v1 engine; no current page names any model provider.

> It uses a language model (GPT-4) as a reasoning engine at key steps in a structured exploration process.
> — https://www.undermind.ai/whitepaper.pdf · Whitepaper 'Benchmarking the Undermind Search Assistant', §1 'Building an ideal search engine for scientists', page 2

**Model choice** · No — The vendor assigns models by plan rather than letting the buyer choose: Free gets strong models, Pro gets the latest ones, and no model picker is documented anywhere.

> The latest, most powerful AI models
> — https://www.undermind.ai/#pricing · Pricing section, Pro plan feature list, first bullet under 'Everything in Free plus:'

_Not established: Open models_

## Where you use it

**In your editor** · Yes — Documented MCP setup covers Cursor, VS Code and GitHub Copilot in VS Code, so the research tools run inside the editor you already use.

> This adds Undermind to VS Code as an MCP server.
> — https://www.undermind.ai/mcp · Connect your agents page, VS Code setup section

**On the command line** · Partly — There is no Undermind CLI of its own; terminal use comes through agent CLIs such as Codex and Claude Code, which you configure with the MCP endpoint.

> codex mcp add undermind --url https://mcp.undermind.ai/mcp --oauth-client-id codex
> — https://www.undermind.ai/mcp · Connect your agents page, Codex setup section, terminal commands

**In a browser** · Yes — The primary product is a browser web app that organisations can deploy to teams without any engineering work.

> Easily deploy Undermind's AI-based literature research capabilities to your teams, available right in the web browser.
> — https://www.undermind.ai/enterprise · Undermind for organizations section, Web app card

## Whose machine it runs on

**Self-hosted** · No — Every delivery shape on the site is Undermind-hosted: a web app, a hosted MCP endpoint and an enterprise cloud with its own access controls; no bring-your-own-infrastructure option is offered.

> Secure Cloud Encryption in transit and at rest, with strict access controls. Trusted by dozens of leading research companies and verified by independent penetration testing.
> — https://www.undermind.ai/enterprise · For enterprises: the security and scale you expect section, Secure Cloud card

**Open source** · No — Closed source: the Terms forbid derivative works and reverse engineering, and no repository, licence or source distribution is published anywhere on the site.

> you shall not modify, make derivative works of, disassemble, reverse compile or reverse engineer any part of the Site
> — https://www.undermind.ai/terms · Terms of Use §2.2 Certain Restrictions, clause (b)

## What it costs to run

**How it meters** · Yes — Billing is per person per month on a plan, with compute rather than credits as the differentiator between Free, Pro and Enterprise tiers.

> per person per month, billed annually
> — https://www.undermind.ai/#pricing · Pricing section, Team plan price cell subtitle (Academic tab, Annual)

**Free tier** · Yes — A standing $0 plan exists with self-serve signup, no sales call, though it searches abstracts and metadata rather than full texts.

> Free For exploratory research $0
> — https://www.undermind.ai/#pricing · Pricing section, Free plan card price cell

**API access** · Partly — An Undermind API with programmatic queries and batch processing is offered to organisations via sales; no public API documentation or self-serve key is published.

> Programmatic queries via API
> — https://www.undermind.ai/enterprise · Undermind for organizations section, Undermind API card, first bullet

**MCP server** · Yes — A hosted MCP server at mcp.undermind.ai exposes deep search, PDF reading and workspace tools, and the Free plan row already includes agent connection.

> Yes. Undermind provides an MCP server, so you can connect the AI assistants and coding agents you already use.
> — https://www.undermind.ai/ · FAQ, answer to 'Can I connect my own agents to Undermind?'

**Bring your own key** · No — No tier offers your own model keys, your own inference or your own cloud; the top tier's published differentiators are compute, org login, onboarding, admin, terms, pricing and support.

> Increased compute Sitewide organizational login Onboarding seminars Admin dashboard Custom terms/SLA/security review Custom pricing Dedicated support
> — https://www.undermind.ai/#pricing · Pricing section, Enterprise plan card, 'All features in Team, plus:' list

## Buying it for a team

**A company can buy it** · Yes — A Team plan with centralised billing is purchasable directly, and Enterprise agreements with custom terms are offered above it.

> Centralized billing
> — https://www.undermind.ai/#pricing · Pricing section, Team plan feature list under 'All features in Pro, plus:'

**Seat model** · Yes — Team is priced per person per month with a seat-count selector on the card, and the published rate is the same whether you buy Pro or Team seats at academic annual pricing.

> per person per month, billed annually
> — https://www.undermind.ai/#pricing · Pricing section, Team plan price cell subtitle (Academic tab, Annual)

**Pooled budget** · No — There is no shared balance to pool: entitlement is a per-user rate limit on searches and requests, not a team allowance.

> Rate limits may apply to the use of our services, restricting the number of searches or requests per user within a given timeframe to ensure fair usage.
> — https://www.undermind.ai/terms · Terms of Use §3.4 Acceptable Use Policy

**Admin controls** · Partly — Enterprise adds an admin dashboard and member management on Team, but no page describes constraining what the agent itself may do.

> Admin dashboard
> — https://www.undermind.ai/#pricing · Pricing section, Enterprise plan feature list under 'All features in Team, plus:'

**Audit log** · Partly — Admins receive usage reports on member accounts, but no audit log of administrative action appears in the Enterprise feature list or the security overview.

> may receive reports regarding use of the Site by those Accounts, and may use any other administrative controls the applicable plan provides.
> — https://www.undermind.ai/terms · Terms of Use §1.4 Organization Administration

**Single sign-on** · Partly — Enterprise offers sitewide organisational login and says an existing enterprise IdP can stay the sign-in path, but SAML and OIDC are never named.

> SSO with your existing enterprise IdP
> — https://www.undermind.ai/mcp · Enterprise integration section, 'Works with the agent stack you already run' checklist

## What happens to your code

**Opt out of training** · Partly — A no-training guarantee is stated for enterprise R&D data, covering model providers too, but the vendor makes no equivalent statement for Free, Pro or Team users.

> Your company's proprietary R&D data stays isolated, is never used to train AI models, and our model providers never retain or train on it.
> — https://www.undermind.ai/enterprise · For enterprises section, No Training card

**Data residency** · Yes — United States storage, with SCCs for EEA/UK/Swiss transfers; no choice of region is offered.

> We process and store personal data in the United States.
> — https://www.undermind.ai/privacy · Privacy Policy, 'International data transfers', first sentence

**Getting out** · Partly — You own inputs and outputs and can demand complete deletion at any time, but no bulk export path is documented and workspace files are only reachable through the app or MCP tools.

> You own your inputs and outputs, we provide the software. Request complete removal of your organization's data at any time.
> — https://www.undermind.ai/enterprise · For enterprises section, Own Your IP card

**Certifications** · Partly — No SOC 2, ISO 27001 or GDPR certification is claimed; the published assurance is third-party penetration testing with a letter of attestation under NDA.

> Documentation: Letter of Attestation / pentest report available under NDA.
> — https://www.undermind.ai/security-one-pager.pdf · Page 1, Pentest Facts, 'Contact & Next Steps'

## Provenance

Every fact above comes from the vendor. Nothing here is independently corroborated yet.

## Coverage elsewhere

Pages about this product that the vendor does not control. The label states how independent each one is.

| Source | Kind | Read |
| --- | --- | --- |
| pmc.ncbi.nlm.nih.gov | Independent source | 2026-08-23 |

## What changed

- 2026-09-26 — Capabilities · Commercial terms · Free tier
- 2026-09-03 — Commercial terms · Free tier · Price
- 2026-08-31 — Capabilities · Commercial terms · Free tier · Price

## Correcting this record

Anything marked "Not established" is a question we have not answered — it is not a No. If you represent Undermind, or anything here is wrong or out of date, you can supply it with a source: the operations this site accepts from an agent are listed at https://theagentsindex.com/.well-known/agents.json — that catalogue needs no key to read, and each entry states its own auth.

Supplying evidence is free. Nothing that can be bought here changes what a field says, whether a gap is closed, or how this record is scored.