Build an AI Stock Research Workflow with Gemini CLI and PlusE

Equity research is often less about one brilliant model and more about moving through the same information pipeline again and again: collect company data, inspect price action, check technical indicators, read news, compare market context, build charts, and summarize the investment case.

That workflow is a good fit for AI agents, but only if the agent can access reliable financial data in a format it can actually use. A generic chatbot with copied screenshots and pasted tables is not enough. The agent needs a connected research environment.

This guide shows one practical setup using Gemini CLI as the agent interface, PlusE MCP as the financial data layer, and a chart MCP server for visualization.

Workflow Overview

The setup has three parts:

1. Agent interface: Gemini CLI handles the research instructions, tool calls, report writing, and follow-up revisions.
2. Financial data layer: PlusE MCP gives the agent access to company profiles, price history, technical summaries, news, sentiment, macro context, options data, and point-in-time snapshots.
3. Visualization layer: A chart MCP server, such as AntV Chart MCP, helps generate charts for review decks and research notes.

The result is not a fully automated trading system. It is a research assistant that can collect data, organize evidence, produce an initial report, and respond to analyst feedback.

Why Data Access Matters

Many AI stock research workflows fail at the data layer. If the agent receives raw daily price rows, fragmented API responses, and unprocessed filings, most of the context window gets spent parsing data instead of reasoning about the investment question.

PlusE is useful here because it exposes financial information in LLM-friendly formats. Instead of forcing the model to derive everything from raw tables, the agent can request prepared summaries and structured market context.

Useful examples include:

• Price summaries: Historical price trends, YTD changes, volatility, drawdowns, and risk metrics without sending every daily candle to the model.
• Technical analysis summaries: RSI, MACD, moving averages, trend signals, and short-term setup context expressed in a form the agent can explain.
• Machine learning predictions: Probabilistic price and volume signals that can be used as one input in a broader research process.
• News and sentiment: Market headlines, company-specific news, and social context for understanding what investors are reacting to.
• Point-in-time snapshots: Historical research context that avoids future data leakage when evaluating past decisions.

This makes the MCP server act more like an intelligence hub than a simple price feed.

Example Research Flow

A typical workflow for researching a public company can look like this:

1. Start Gemini CLI in the project workspace.
2. Connect PlusE MCP as the stock market data source.
3. Connect a chart MCP server for visual outputs.
4. Ask the agent to research a ticker with a structured prompt.
5. Review the first report and ask follow-up questions.
6. Export the final report, charts, and key assumptions.

For example, an analyst researching Walmart could ask the agent to gather company fundamentals, recent price behavior, technical setup, macro context, recent headlines, and relevant risk factors. The agent can then call the connected tools, summarize the findings, create charts, and draft a report.

The key is to avoid vague prompts like "research Walmart." Better results come from giving the agent a clear research mandate.

Prompt Template

Use a structured prompt like this as a starting point:

Research {TICKER} as an equity analyst.

Goal:
Produce a concise investment research memo for a human analyst to review.

Required sections:
1. Executive summary
2. Company overview
3. Recent price action and technical setup
4. Fundamental snapshot
5. News and sentiment context
6. Macro or sector factors
7. Bull case
8. Bear case
9. Key risks
10. Watchlist levels and follow-up questions

Data requirements:
- Use PlusE for market data, financial data, technical context, news, and sentiment.
- Prefer summarized or LLM-ready outputs when available.
- Use point-in-time context if I provide an as-of date.
- Create charts only when they clarify the thesis.

Important constraints:
- Do not invent missing data.
- Clearly separate facts, model output, and interpretation.
- Treat predictions as signals, not as guarantees.
- End with questions a human analyst should verify before making a decision.

Setup Steps

1. Install Gemini CLI

Install and authenticate Gemini CLI using Google's official instructions. After it is available in your terminal, create or open a workspace for your research project.

2. Add PlusE MCP

Create a PlusE account, get your MCP connection details, and add the MCP server to Gemini CLI.

The exact command depends on your local Gemini CLI configuration, but the flow is:

gemini mcp add pluse

If you need the current setup instructions, see the MCP client setup guide.

3. Add Chart Generation

For visual reports, connect a chart MCP server such as AntV Chart MCP. This lets the agent create price charts, trend comparisons, allocation charts, and other visuals during the research process.

Charts are especially useful for:

• Explaining price trends
• Comparing revenue or margin changes
• Showing valuation scenarios
• Summarizing sector or portfolio exposure

4. Run the Research Prompt

Once the MCP servers are connected, run Gemini CLI and paste a structured prompt. Start with a single ticker and a focused question.

Examples:

• "Research WMT for a 3-month swing trading setup."
• "Compare META and GOOGL from a long-term valuation perspective."
• "Create a point-in-time research memo for AAPL as of 2023-03-01."

After the first draft, ask the agent to tighten weak sections, add missing evidence, or generate charts for specific claims.

What This Improves

This workflow can reduce the most repetitive parts of equity research:

• Manual data collection across multiple sites
• Copying tables into spreadsheets or chat windows
• Recalculating common indicators
• Summarizing headlines one by one
• Creating first-draft research notes from scratch
• Reworking charts for internal review

The human analyst still owns the conclusion. The agent accelerates the data collection, context building, and first-draft synthesis.

Best Practices

• Keep prompts specific. Define the ticker, time horizon, investment style, and required sections.
• Ask for assumptions. Require the agent to state what data it used and where confidence is low.
• Separate signals from decisions. Technical summaries and model predictions should support the thesis, not replace judgment.
• Use point-in-time mode for backtests. Avoid future leakage when researching historical scenarios.
• Review every output. Treat AI-generated research as a draft that needs human validation.

Reference Project

An example configuration and prompt collection is available here:

https://github.com/wanghsinche/vibe-trading

Use it as a starting point for building your own AI-assisted stock research workspace.