Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@QuantContext Screen S&P 500 for value stocks with PE under 15 and ROE above 12%"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
QuantContext is an MCP server that turns plain-English strategy descriptions into executable quant research: screen stocks by any criteria, backtest over historical data, and run factor analysis to see where the returns come from. Every number is computed from real market data, not generated by an LLM. Results are fully reproducible.
Works with Claude, Codex, OpenCode, or any other MCP-compatible coding agent.
pip install quantcontext-mcpClaude Code:
claude mcp add quantcontext -- quantcontextClaude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"quantcontext": {
"command": "quantcontext"
}
}
}No API keys. No configuration.
Related MCP server: panther-mcp
Three tools that compose into a full research workflow:
screen_stocks -> backtest_strategy -> factor_analysisTool | What it does |
| Filter S&P 500, Nasdaq 100, or Russell 2000 by fundamentals, momentum, quality, technical signals, or a multi-factor blend. Returns ranked candidates. |
| Test a strategy over history with a rebalance-loop engine. Returns CAGR, Sharpe, max drawdown, equity curve, and trade log. |
| Decompose strategy returns into Fama-French factors (market, size, value, momentum). Returns alpha with t-statistic, factor loadings, and R-squared. |
Stock screening:
Screen S&P 500 for value stocks: PE under 15, ROE above 12%Find the top 20% momentum stocks in the Nasdaq 100 over the last 200 daysRank S&P 500 stocks by a blend of value, momentum, and quality, equal weight each factorFind S&P 500 stocks with RSI under 40 and price above the 200-day moving averageBacktesting:
Backtest a top-20% momentum strategy on Nasdaq 100, monthly rebalance, last 2 yearsHow would a value screen (PE under 15, ROE above 12%) have performed on S&P 500 over the last 3 years?Test a momentum strategy with a 15% stop loss and 20% max portfolio drawdown circuit breakerFull research workflow:
Screen S&P 500 for cheap, high-quality stocks. Backtest monthly over 3 years,
then run factor analysis. Is the return real alpha or just factor exposure?Screen | Description | Key parameters |
| Filter by PE, ROE, leverage, revenue growth |
|
| Profitability and balance sheet health |
|
| Rank by N-day price momentum |
|
| Cheapest stocks by valuation |
|
| Multi-factor composite score |
|
| RSI and SMA crossover signals |
|
| Stocks below z-score threshold |
|
The tools are also importable directly — no agent required. Useful if you have an existing script and want to plug in backtesting or factor analysis.
from quantcontext.server import screen_stocks, backtest_strategy, factor_analysis
import asyncio, json
# Screen
result = json.loads(asyncio.run(screen_stocks(
universe="sp500",
screen_type="fundamental_screen",
config={"pe_lt": 15, "roe_gt": 12},
)))
# Backtest
bt = json.loads(asyncio.run(backtest_strategy(
stages=[{"order": 1, "type": "screen", "skill": "fundamental_screen", "config": {"pe_lt": 15, "roe_gt": 12}}],
universe="sp500",
rebalance="monthly",
start_date="2022-01-01",
)))
print(bt["metrics"])
# Factor analysis — pipe the equity curve straight in
fa = json.loads(asyncio.run(factor_analysis(
equity_curve=bt["full_equity_curve"]
)))
print(fa["alpha_annualized"], fa["alpha_tstat"])Strategies are expressed using the built-in screen types from the table above. All functions are async and return JSON strings.
All public data, no API keys required.
Data | Source | Cache |
Daily OHLCV prices | Yahoo Finance ( |
|
Fundamentals (PE, ROE, margins, etc.) | Yahoo Finance |
|
Fama-French factors (Mkt-RF, SMB, HML, Mom) | Kenneth French Data Library |
|
Universe lists (S&P 500, Nasdaq 100) | Wikipedia |
|
The first tool call downloads and caches data (10-30 seconds). All subsequent calls use the local cache: screening under 1s, backtesting 3-8s.
To skip the cold start, run once after install:
quantcontext-warmup --url https://quantcontext.ai/api/dataMIT
Run a historical backtest on a stock screening strategy. Uses a rebalance-loop engine that re-runs the screening pipeline on each rebalance date, sizes positions, enforces risk limits, and tracks daily P&L.
Returns equity curve, trade log, and performance metrics including CAGR, Sharpe ratio, maximum drawdown, Calmar ratio, win rate, and turnover.
The backtest is fully deterministic — same inputs always produce identical results.
After backtesting, use factor_analysis on the equity_curve to decompose returns into Fama-French factors (market, size, value, momentum) and estimate true alpha.
| Name | Required | Description | Default |
|---|---|---|---|
| stages | Yes | Pipeline stages defining the strategy. Each stage is an object with: order (int), type ('screen'|'analyze'|'signal'), skill (skill name), config (dict). Example: [{order: 1, type: 'screen', skill: 'fundamental_screen', config: {pe_lt: 15}}, {order: 2, type: 'signal', skill: 'momentum_screen', config: {lookback_days: 200, top_pct: 0.3}}] | |
| universe | No | Stock universe. Options: sp500, russell2000, nasdaq100 | sp500 |
| rebalance | No | Rebalance frequency. Options: daily, weekly, monthly, quarterly | monthly |
| sizing | No | Position sizing method. Options: equal_weight, inverse_volatility | equal_weight |
| start_date | No | Backtest start date in YYYY-MM-DD format | 2023-01-01 |
| end_date | No | Backtest end date in YYYY-MM-DD format. Defaults to today. | |
| max_position_size | No | Maximum weight per position (0-1). E.g., 0.1 = 10% max per stock | |
| stop_loss | No | Per-position stop loss (0-1). E.g., 0.15 = sell if position drops 15% | |
| max_drawdown | No | Maximum portfolio drawdown before going to cash (0-1). E.g., 0.2 = 20% |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description adds value beyond annotations by stating the backtest is fully deterministic (idempotent) and lists return metrics. It aligns with readOnlyHint and destructiveHint, with no contradictions.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is concise (5 sentences) with a clear front-loaded purpose. Each sentence adds distinct value, including deterministic behavior, outputs, and follow-up guidance.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool's complexity (9 params) and presence of an output schema, the description covers the core purpose, return types, and behavior adequately. It could mention more about the rebalance-loop engine details, but is still reasonably complete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so the description adds no extra parameter meaning beyond what the schema already provides. The description only mentions return values, not parameter details, so baseline 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool runs a historical backtest on a stock screening strategy, with specific outputs (equity curve, trade log, performance metrics) and mentions a rebalance-loop engine. It distinguishes from sibling tools by suggesting factor_analysis for subsequent use.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explains when to use the tool (for backtesting) and provides a follow-up step (use factor_analysis). However, it does not explicitly state when not to use it or compare to alternatives like screen_stocks, though the sibling context implicitly differentiates.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Decompose strategy or portfolio returns into Fama-French factors using OLS regression.
Breaks down returns into exposures to four systematic factors:
Mkt-RF (market risk premium): how much return comes from overall market movement
SMB (small minus big): size factor exposure
HML (high minus low): value factor exposure
Mom (momentum): momentum factor exposure
Also estimates alpha (excess return not explained by factors) with t-statistic for statistical significance. A |t-stat| > 2 suggests statistically significant alpha.
Returns alpha (daily and annualized), factor loadings with t-statistics, R-squared (how much of return variance is explained by factors), and residual volatility.
Use this after backtest_strategy to understand WHERE your returns come from — is it genuine alpha or just factor exposure?
| Name | Required | Description | Default |
|---|---|---|---|
| equity_curve | Yes | Equity curve as a list of {date, value} objects. Typically from the output of backtest_strategy. Needs at least 30 data points. Example: [{date: '2023-01-03', value: 100000}, {date: '2023-01-04', value: 100500}, ...] |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already convey readOnlyHint and non-destructive nature. Description adds value by detailing the statistical interpretation (alpha with t-stat significance threshold, R-squared, residual volatility) and listing the four factors. No contradictions.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Description is concise yet thorough, using bullet points for factors and clear paragraphs. Front-loaded with main purpose, then details, then usage guidance. Every sentence adds value.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the output schema exists (not shown but known from context), description doesn't need to itemize return values. It covers prerequisites, interpretation guidance (t-stat cutoff), and outputs conceptually, making it complete for the tool's complexity.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100% with a detailed description for the single parameter. The description provides additional context by noting the parameter typically comes from backtest_strategy output and including an example. Baseline 3 is appropriate as schema already does heavy lifting.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description begins with a specific verb ('Decompose') and resource ('strategy or portfolio returns into Fama-French factors'), clearly stating what the tool does. It distinguishes from siblings by explicitly recommending usage 'after backtest_strategy to understand WHERE your returns come from — is it genuine alpha or just factor exposure?'
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly states when to use ('after backtest_strategy'), and provides conditions ('Needs at least 30 data points'). No explicit exclusions for alternatives, but context clarifies its purpose relative to sibling tools.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Screen a stock universe with quantitative filters. Returns ranked candidates with scores and metrics.
Use this tool when you need to find stocks matching specific criteria — value stocks, momentum leaders, quality companies, or multi-factor ranked candidates. Supports 7 screen types across 3 universes (S&P 500, Russell 2000, Nasdaq 100).
After screening, use backtest_strategy to test the screen as a trading strategy, or factor_analysis to understand the factor exposures of the selected stocks.
| Name | Required | Description | Default |
|---|---|---|---|
| universe | No | Stock universe to screen. Options: sp500, russell2000, nasdaq100 | sp500 |
| screen_type | No | Type of screen to run. Options: fundamental_screen (filter by PE/ROE/debt), quality_screen (filter by ROE/margins), momentum_screen (rank by price momentum), value_screen (rank by valuation), factor_model (multi-factor ranking), technical_signal (RSI/SMA/Bollinger), mean_reversion (z-score below threshold) | fundamental_screen |
| config | No | Screen-specific configuration. Examples: fundamental_screen: {pe_lt: 15, roe_gt: 12}. momentum_screen: {lookback_days: 200, top_pct: 0.2}. value_screen: {pe_lt: 20, top_n: 30}. factor_model: {weights: {value: 0.3, momentum: 0.3, quality: 0.2, volatility: 0.2}, top_n: 20}. mean_reversion: {lookback_days: 60, z_threshold: -1.5}. All parameters are optional — sensible defaults are used. | |
| date | No | Date for the screen in YYYY-MM-DD format. Defaults to most recent trading day. |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already show read-only and idempotent behavior. Description adds context about supporting 7 screen types across 3 universes and that parameters have sensible defaults, which helps the agent understand non-obvious traits.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Concise yet informative: first sentence states purpose, then usage guidance, then examples, then post-use hints. No unnecessary words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Despite having 4 parameters and a complex config object, the description covers all aspects: parameter defaults, examples, universe options, screen types, and next-step tools. Output schema exists (not shown but referenced), so completeness is high.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100% with descriptions for all parameters. The description enhances this by providing example configs for each screen type, adding meaning beyond schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Clearly states it screens stocks with quantitative filters and returns ranked candidates. Distinguishes from siblings (backtest_strategy, factor_analysis) by mentioning them as post-use tools.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly tells when to use this tool ('find stocks matching specific criteria') and what to do after screening. Provides context on screen types and universes.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.2.0backtest_strategyfactor_analysisscreen_stocksScored across 3 tools
Each tool has a distinct purpose: screen_stocks finds candidates, backtest_strategy tests strategies, factor_analysis decomposes returns. No overlap in functionality.
All tool names follow a consistent verb_noun pattern in snake_case: backtest_strategy, factor_analysis, screen_stocks.
Three tools is minimal but well-scoped for the quantitative finance pipeline. Covers screening, backtesting, and analysis without excess.
The tools form a coherent workflow (screen → backtest → factor analysis). Minor gaps exist, such as no data retrieval or custom factor tools, but the core pipeline is complete.
Build, backtest, and deploy quantitative trading strategies from your AI agent.
Analytical engine for US-listed equities: screens, rankings, and event studies in plain English.
Backtest plain-English trading strategies on real market data: graded results, honesty flags.
Quant intelligence over MCP: backtest, signals, screens, scores & portfolios for US & TSX stocks.
