Repository avatar
Other Tools
v0.1.0
active

openlandscape

io.github.theopenlabs/openlandscape

Drug pipelines from public trial registries and regulators: every drug for a disease, with sources.

Documentation

OpenLandscape: every drug in development, with a source for every fact. Free to start, $49 a month, open source.

Drug-pipeline landscapes from public registries: every drug for a disease, target or company,
who owns it, how far it has got and what happens next. On the web, and inside your AI assistant.

openlandscape.app · Try it (no account) · Use it from your AI · vs Citeline · What it costs · Run it yourself

Why OpenLandscape

A source for every fact. Every stage, date and number links to where it came from: ClinicalTrials.gov, the EU trials register, FDA, EMA, SEC filings or the company's own news. Nothing to take on trust.

A fair price. Pipeline databases such as Citeline are sold to big pharma for tens of thousands a year, after a sales call. OpenLandscape starts free and costs $49 a month for everything.

Open by default. The code is here, so anyone can see how every list is made. Your AI assistant can use it directly through the MCP connector.

What it does

A landscape for any disease, target or company

Search for an area and get every drug in active trials: how it works, who owns it, its stage and its next date. Filter by stage, mechanism or China-based companies, copy the table or download it. Ready-made landscapes (obesity, PD-1/VEGF bispecifics, in vivo CAR-T, lung cancer) are free to read without an account.

The obesity landscape: what changed this week, what is coming up, and every drug with how it works, company, stage and next date

Every drug, with its trials and sources

A drug page says what the drug is in plain words, its other names and code names, every trial and what it is tested for, with the registry entry behind each line.

The mazdutide page: stage, next date, active trials and targets, and what the drug is, in plain words with sources

Every company's pipeline

A company page lists everything it runs in trials, what is furthest along, what reads out next, and its SEC filings.

The Novo Nordisk page: drugs furthest along, next date, number of drugs and trials, and the drug list

Ask your AI

Add the connector to Claude, ChatGPT, Cursor or any MCP app and ask in plain words. The answer comes from the same data, with the sources attached.

A real answer through the connector: oral GLP-1 pills in Phase 3 or later, with company, stage and next date

Follow an area, hear what changed

Follow a landscape and see every new trial, moved date, approval and filing as it happens. A Monday email sums up the week and the dates coming up.

The Monday email: what changed this week and what is coming up, each line with its source

Light, dark and on your phone

The PD-1/VEGF bispecifics landscape in dark mode

OpenLandscape on two phones: the obesity landscape and the mazdutide page

Screenshots are from the live data. The account in them (alex@example.com) is made up.

OpenLandscape and the big pipeline databases

OpenLandscapeCiteline and similar
PriceFree to start, then $49 a monthTens of thousands a year
Getting startedSign in with your email, no cardA sales call and a contract
Where the facts come fromPublic registries, regulators, filings and company news, each linkedAlso their own analysts and private research
CodeOpen source (AGPL-3.0)Closed

The big databases cover more, because they pay analysts to read what is not public. OpenLandscape covers what is public, shows you the source for each line, and costs a small part of the price.

What it costs

  • Free: no card, no time limit. The top 10 of any list, one landscape in full and 25 AI questions a month.
  • Solo: $49 a month or $490 a year (before VAT), for one person. Every list in full, follows, every change as it happens and the Monday email.

Details: openlandscape.app/pricing.

Use it from your AI assistant

OpenLandscape is a remote MCP server (Streamable HTTP). Server URL:

https://openlandscape.app/mcp

You sign in with your OpenLandscape account (OAuth) the first time; a free account works.

  • Claude (app or claude.ai): find OpenLandscape in the connector directory, or Settings → Connectors → Add custom connector, name it OpenLandscape and paste the URL.

  • Claude Code:

    claude mcp add --transport http openlandscape https://openlandscape.app/mcp
    
  • Cursor: add to ~/.cursor/mcp.json:

    { "mcpServers": { "openlandscape": { "url": "https://openlandscape.app/mcp" } } }
    
  • Gemini CLI: add to ~/.gemini/settings.json, restart, and type /mcp to check:

    { "mcpServers": { "openlandscape": { "httpUrl": "https://openlandscape.app/mcp" } } }
    
  • ChatGPT and other apps: step-by-step on openlandscape.app/connect.

Then ask, for example: "Map every amylin drug in Phase 2 or later, with next milestones."

Tools: search, landscape, trials, catalysts, who_runs_trials, drug, changes, whoami, and show (draws the answer as a card in apps that support it). Every answer carries its sources.

More for AI tools: openlandscape.app/ai and openlandscape.app/llms.txt.

Run it yourself

What is in this copy, and what is not

This repository is the open-source copy of the code. It has everything the site runs on: the data loaders, the grouping of trials into drugs, the website, sign-in, billing, the AI connector and the checks.

It does not have the checked data. People check the rows by hand: which trial belongs to which drug, which company owns what, which names mean the same drug. That work is what OpenLandscape sells, so it stays private. Also private: company logos and the list of company aliases (for example that "GSK" means GlaxoSmithKline).

So a copy you build yourself is thinner than the live site:

  • One checked area: obesity (seed/obesity.js), as a sample of what the checked data looks like.
  • Every other area is grouped automatically from the registry names. It works, but expect more duplicates and more drugs without a known company or target.
  • No company logos. Some company searches by short name find nothing.
  • The files that hold checked data are empty stand-ins with the right shape (in seed/), so the code runs.

Run it on your computer

Needs Node 22 or later, and about 1 GB of free disk space.

npm install
npm run ingest      # downloads trials from ClinicalTrials.gov and approvals from openFDA, builds the database (15 to 20 minutes)
npm start           # http://localhost:8080

Then open localhost:8080/landscapes/obesity.

The database is one SQLite file, data/openpipeline.db (a few hundred MB). npm run ingest builds it from public sources only. Most of the time goes on looking up drug names in public drug lists (the FDA substance registry and the NCI drug dictionary) and FDA labels. The drug-list lookups stop after a few minutes, so some names are not asked, and a few rows that are the same drug stay apart. Run npm run ingest again to refresh.

No keys are needed to run it locally:

  • Without an email service, sign-in links are printed in the server log and shown on the page.
  • Without a Stripe key, the pricing page shows, but checkout says payments are not configured.

To add keys, copy .env.example to .env and fill in what you need. SETUP.md lists every account and key, and how to run your own server.

Run the checks

TODAY=2026-11-06 npm run check

TODAY fixes the date, so date-based checks give the same answer on any day. Some checks need the private checked data. In this copy they print SKIP and do not fail. That is expected.

How it works

PieceWhat it doesWhere
Trial loaderDownloads industry drug trials (phase 1 to 4) from the ClinicalTrials.gov APIsrc/ingest/ctgov.js
Approval loaderDownloads the FDA's list of approved drugs (openFDA)src/ingest/openfda.js
Checked dataHand-checked companies, drug names, approvals and filings, with sourcesseed/ (obesity only here)
GroupingGroups trials into drugs, merges names for the same drug (mazdutide = LY3305677 = IBI362), skips placebo and comparator armssrc/ingest/link.js
QuestionsSearch, landscape, trials, catalysts, who runs trials, one drug, recent changessrc/query.js
AI connectorThe same questions for AI apps, as a remote MCP serversrc/mcp.js
WebsiteLandscape pages, dashboard, pricing, Stripe checkout, email sign-insrc/server.js, src/pages/, web/
ClientThe connector setup for Claude and ChatGPT, and skills for writing landscape memosclient/

Checked areas are listed in seed/index.js. To check a new area, add a seed file in the same format as seed/obesity.js and list it there.

Why the ClinicalTrials.gov API and not AACT? AACT is a nightly database copy of the same registry. It needs an account, and we only need the industry drug trials, a small part of it. The official API serves the same records with no account.

The AI connector

The connector is a remote MCP server at /mcp. AI apps sign in with OAuth (src/oauth.js); /mcp/u/<api-key> also works for apps that take a plain address. It has eight tools:

whoami, search, landscape, trials, catalysts, who_runs_trials, drug, changes.

A ninth tool, show, draws the answer as a card inside Claude or ChatGPT. Set AI_CARDS=0 in .env to turn cards off; show then goes away and the tools answer in text only.

Rules the product follows

  1. Every phase, date and number links to its source.
  2. Facts from press reports carry a "news" tag. Name merges and mechanisms not yet confirmed carry an orange dot.
  3. Every page says when the data was pulled.
  4. Features not built yet show the month they ship.

Contributing and security

Licence

  • The app (server, data loaders, AI connector, website): AGPL-3.0, see LICENSE.
  • The client (client/): MIT, see client/LICENSE.

The data the app downloads comes from ClinicalTrials.gov, openFDA and SEC EDGAR, under their own terms.

No installation packages available.

Remote