Ask in a chat, and the assistant reads QuanterLab's published research and your own saved work, and runs backtests, walk-forwards and scans on your account. You need a QuanterLab account, and a free one works: it reads everything on every plan and runs three a day on the Free plan (what each plan runs). Sign up for free.
ChatGPT
QuanterLab is in ChatGPT's plugin directory, free ChatGPT accounts included. Setup in three steps.
@QuanterLab and ask. Try: What did QuanterLab find about dividend capture? Give the headline numbers and the link to the paper.To add it by hand instead, in ChatGPT's Developer mode, the server URL is https://quanterlab.com/mcp, with OAuth.
ChatGPT calls an MCP server a plugin. It connects through OAuth, which this server speaks (dynamic client registration, PKCE), so the same URL connects. It also asks every plugin for two tools of its own, search and fetch, which here cover the published research library: a search returns papers with their public URLs, a fetch returns a paper's text, so ChatGPT can cite a QuanterLab paper by its page. Your own work stays behind the list tools below, under your login.
Claude
QuanterLab is in Claude's plugin directory, for claude.ai, the Claude desktop app, Cowork and Claude Code. The plugin brings this connector and the skills that tell Claude how to use it. Plugins come with Claude's paid plans (Pro, Max, Team and Enterprise); on Claude's Free plan, add the connector by hand instead (below). Setup in three steps.
This route works on every Claude plan, Free included (Claude's Free plan allows one custom connector; on Team and Enterprise an Owner adds it for the organisation). In claude.ai's connector settings, use this URL:
https://quanterlab.com/mcp filled in (the URL is on your clipboard too); confirm it.In either assistant
Once connected, ask things like
What each plan runs.
| Free | Practitioner | Research | |
|---|---|---|---|
| Read the studies and your saved work | Yes | Yes | Yes |
| Runs a day | 3 in all | Each kind up to its cap | Each kind up to its cap |
| Runs at the same time | 3 | 3 | 3 |
A run is a backtest, a walk-forward, a scan, a regime analysis or one of your saved lab circuits, on daily bars. Each kind's daily cap, the same on every plan: 50 backtests, 50 stochastic-method backtests, 10 walk-forwards, 10 lab circuits, 30 mean-reversion scans, 15 pairs scans, 20 regime analyses and 10 regime optimisations. A lab circuit with a Research card in it (the factor models, the routers, paper-trade markers) runs on Research, as in the lab. A refused call does not count, and a run outside your plan or past a cap is refused with a message that says why. The plans.
Three surfaces
Skills
A skill is a short file an assistant reads before it acts. These four state, in the order a model needs them, what keeps a study on QuanterLab clean: the traps, the structure of the work, the shape of a status report, and the bones of a paper. They are written for Claude Code and Claude Desktop; any assistant that reads instructions can use them.
Install. QuanterLab Desktop installs the four for the assistant seated in it; nothing to do there. Claude Code: unzip into ~/.claude/skills/ for every project, or into a project's .claude/skills/ for that project alone; each skill then answers to /quanterlab-research-hygiene and the other three names. Claude Desktop: Settings, Skills, add each SKILL.md. The desktop app's own MCP server also serves three prompts, study_setup, study_hygiene and paper_prep, which appear as slash commands in Claude Code and as templates in Claude Desktop without any installation.
Reference
Claude picks the tool from plain language; the list is for readers who want to see what is there. Each tool that starts a run says which plan runs it, and each run returns a task that the progress tools follow and summarize_workflow reads.
start_backtest, start_walkforward: each start is a run (what each plan runs).list_indicators: the indicator catalogue with parameter shapes.get_backtest_progress, get_walkforward_progress: progress of a run.start_stochastic_backtest, start_mr_scan, start_pairs_scan: each start is a run (what each plan runs).list_stochastic_methods: the method catalogue.get_stochastic_backtest_progress, get_mr_scan_progress, get_pairs_scan_progress: progress of a run.start_regime_analysis, start_regime_optimization: each start is a run (what each plan runs).get_regime_analysis_progress, get_regime_optimization_progress: progress of a run.start_primitives_run: runs one of your saved canvas projects; each start is a run (what each plan runs), and one with a Research card runs on Research.list_primitives: the catalog and what each primitive does.get_primitives_run_progress, get_primitives_run_result: progress and the summarised result of a run.list_my_primitives_projects: your saved circuits and studies.list_fundamental_analyzers: the fundamental analyzers and their inputs.list_my_backtests, list_my_filter_results, list_my_trading_configs, list_my_projects, list_my_running_tasks.save_trading_config, save_workflow_result, summarize_workflow: save and summarise what a session produced.search, fetch: the published papers, with their public links.describe_capability: which workflow fits a question, and what it needs.Claude Desktop, in claude_desktop_config.json:
{
"mcpServers": {
"quanterlab": { "url": "https://quanterlab.com/mcp" }
}
}
Restart Claude Desktop; the sign-in opens in your browser on first connection. Claude Code:
claude mcp add --transport http quanterlab https://quanterlab.com/mcp
Then /mcp inside Claude Code to sign in. Claude in Chrome reads the connector list of your claude.ai account.
For the desktop's local server, the one that runs and registers research, the configuration is written by the desk's Assistant panel with one click; the manual form is on the desktop page.