# AlphaLab > AlphaLab is the AI quant lab behind your agent. Connect it over MCP to the AI agent you already use, like Claude or ChatGPT, and start with a trading hypothesis. Your agent gets the market data, quant tools and research workflows to take it through signal research, backtesting, optimization, robustness and out-of-sample testing. Data covers stocks, futures, forex, options, prediction markets and crypto, plus news and fundamentals, on a deterministic backtesting engine. Access is currently by invitation via a waitlist. You work through your own agent; AlphaLab supplies the data, the engine and the research process. Your agent composes a trading strategy as a directed graph of transformation components and runs it against real market data. AlphaLab handles data, caching, compute, optimization, validation and portfolio simulation, and keeps the research history, so neither you nor your agent has to own the stack. ## What you can do - Build a strategy as a canvas of connected components (data fetch, indicators, signals, transformations, portfolio simulation). - Search tradable instruments across stocks, futures, crypto, and forex. - Run a simple backtest against historical market data. - Run an optimization (discovery) backtest that sweeps parameters with a Bayesian optimizer. - Run an out-of-sample (walk-forward) backtest to validate a strategy on unseen data. - Run a real-time backtest. - Run robustness and validation checks: walk-forward, CPCV, Monte Carlo, regime and cost stress. - Keep research state (projects, hypotheses, notes) in AlphaLab, so you can switch agents or continue in the browser. ## Connecting your agent (MCP) - MCP endpoint: https://api.alphalab.ai/mcp - Auth: OAuth. Add the endpoint to your MCP client and sign in to AlphaLab when it asks. - Start: call get_started, then get_guide for the topics the task needs. - Main workflow: search_instruments -> get_component_contracts -> create_canvas (or load_example_strategy) -> validate_execution_request -> run_experiment -> await_operation -> get_run_metrics. - Validation: start_robustness_analysis, run_validation_suite, get_out_of_sample_windows, get_cpcv_paths. - Research state: create_research_project, create_hypothesis, record_research_note. Work your agent does is saved in AlphaLab and visible in the web app. get_remaining_capacity reports how much research capacity is left for the cycle. There is no need to check it before a run: a run is refused with capacity_exhausted when there is none left, and a smaller run does not help, because the check is account-level rather than per-run. ## Plans The lab is hired by the month, through two doors, firms first. Every plan ships the same staff, the same built desk, the same independent Risk Analyst sign-off and the same cadence; the number of researchers is what changes, and researchers are parallel investigation capacity (how many hypotheses run at once). There is no free plan, and no plan below $500/mo. Firms, engaged through contracts. Multi-seat, with a shared workspace, attribution and sign-off records, weekly status, a monthly review and dossiers: None of these carries a list price. They are contracted rather than checked out, and the researcher count and the terms are agreed with the firm. - Firm Desk: 1 to 5 researchers, each sized for a firm's cadence (2.5x to 7.5x the working capacity of the self-serve desk), so 1 to 5 hypotheses investigated in parallel. At five, a full backlog running at once. There is no menu of firm plans: where a firm lands in that range is scoped in the engagement rather than picked from a list. - Free Pilot: free. One scoped investigation, about thirty days, through the full process to a defended strategy or an honest refutation. This is how a firm relationship starts. - Lab Setup: a fixed term, the first three to six months. For firms and family offices with capital and theses but no quant capability yet: we stand the desk up, turn the implicit rules of the book into a hypothesis backlog, train a designated person through the Mentor and monthly reviews, and graduate them onto a desk they run themselves. - Profit share: by agreement, in place of or alongside a retainer. Proprietary firms trading their own capital only. Self-serve. One named human directing one researcher, on their own book or on one firm account before terms: - Your Desk: $990/mo. 1 researcher. 1 seat. A card-gated 14-day trial is the demo, sized for one researcher to take one hypothesis through validation. A small number of pilots and Lab Setup programs run at a time, as policy, so that high-touch work does not crowd out the firm motion. Contact tarek@alphalab.ai and ask about the current slot. Moving from the self-serve door to a firm engagement carries the backlog, the history and the forward records across intact. Every plan includes every method, unlimited strategies, strategy notes, chart presets, basic and advanced transformations, and MCP API access. Markets (stocks, futures, forex, options, prediction markets, crypto) are available on every plan, as are news and fundamentals. The Risk Analyst is never a paid add-on: charging for the independent check would undermine the results it exists to defend. Its sign-off is a deterministic robustness run on a platform-enforced configuration, never model judgment, and what the team produces is deterministic code executed on the engine (reinforcement learning excepted), so every simulated trade is reproducible. Paper deployment is included; live deployment is a gated state requiring the customer's explicit authorization. ## Concepts - Canvas: a strategy, represented as a directed graph of components. - Component: a single transformation node (for example a data fetch, an indicator, a signal, or a portfolio simulation). - Contract: the typed input/output specification of a component, used to validate connections. - Execution style: simple, optimization (discovery), out-of-sample (walk-forward), or real-time. - Researchers vs seats: researchers are parallel investigation capacity, and the only capacity unit; seats are access and audit only, never charged per unit and never affecting capacity. ## Company AlphaLab is a product by Edgebot (https://www.edgebot.com). Learn more at https://www.alphalab.ai.