User Guide

1. Create a Research Session

Every research project starts with a session. A session groups your companies together and defines how they'll be scored.

1. 1.Go to the Config page
2. 2.Enter a session name (e.g., "Q2 2026 AI Readiness - IT Services")
3. 3.Choose a scoring template (AI Positioning Index or Digital Maturity Index) or upload your own schema
4. 4.Click "Create Session"
5. 5.Your new session will automatically become active in the top bar

2. Add Companies

You have two ways to add companies to your session:

3. Scrape Company Data

Scraping visits each company's website, LinkedIn, and Crunchbase to collect data.

1. 1.After importing, go to the Companies page
2. 2.Click "Scrape" next to individual companies, or go to Upload and click "Start Scraping All"
3. 3.Status changes: pending → scraping → scraped (or failed)
4. 4.Scraping takes 30-60 seconds per company (it visits multiple pages)
5. 5.Check the Activity page to monitor progress and catch errors

4. Review Extracted Data

Once scraped, the system automatically extracts structured data from each company.

1. 1.Click any company name to open its detail page
2. 2.The Dimensions tab shows what was extracted for each scoring dimension
3. 3.Each field has a confidence level: verified, inferred, unverified, or missing
4. 4.The Screenshots tab shows captured page screenshots
5. 5.The Sources tab lists every URL that was visited

5. Score Companies

Scoring rates each company 1-5 on every dimension in your schema.

1. 1.Scoring is done through the MCP server — Claude reads the extracted data and assigns scores
2. 2.Connect Claude to the MCP server (see MCP Setup below)
3. 3.Ask Claude to score companies based on the extracted data
4. 4.Once all dimensions are scored, the company status becomes "analyzed"
5. 5.You can also score manually via the API

6. Export Results

Generate a formatted Excel workbook with all your research.

1. 1.Go to the Export page
2. 2.Optionally enter a filename
3. 3.Click "Generate Export"
4. 4.Download the Excel file — it includes:

• Methodology sheet (scoring framework explanation)
• One sheet per company tier with all extracted data
• Master list with scores and maturity levels
• Sources log with full citations

Session Management

Sessions keep your research organized. Each session has its own scoring schema and company list.

1. 1.Switch between sessions using the dropdown in the top navigation bar
2. 2."All Sessions" shows companies across all sessions
3. 3.Create new sessions from the Config page
4. 4.Each session can use a different scoring template
5. 5.Sessions can be archived (not deleted) when you're done with them

Custom Schemas

You can create your own scoring frameworks beyond the built-in templates.

1. 1.On the Config page, choose "File" or "Text" input mode when creating a session
2. 2.Upload a YAML or JSON file defining your dimensions, fields, keywords, and scoring rubrics
3. 3.Or paste the schema text directly
4. 4.The Config page shows the full schema for your active session — use it as a reference

Data Quality

Monitor the confidence and completeness of your extracted data.

1. 1.See what percentage of fields are verified, inferred, unverified, or missing
2. 2.View quality breakdown by dimension — identifies weak areas
3. 3.Use this to decide which companies need re-scraping or manual review

Activity Log

Track everything that happens in the system.

1. 1.See scrape completions, failures, imports, exports, and score events
2. 2.Filter by level (info/warning/error) or category (scrape/import/export/etc.)
3. 3.Click any row to expand and see full error details
4. 4.Enable auto-refresh to watch events in real time

Connecting Claude to the Research Agent

The MCP server lets Claude read your research data, score companies, and trigger exports — all through natural conversation.

1. 1.The MCP server runs as a local process that Claude connects to via stdio
2. 2.Add this to your Claude MCP configuration:

• query_companies — search and filter companies
• get_company_detail — full detail with extracted data and scores
• get_raw_content — scraped markdown content
• get_extracted_data — parsed fields with confidence levels
• get_sources — citation log
• get_data_quality — confidence metrics
• set_score — rate a company on a dimension (1-5 with rationale)
• list_sessions — see all research sessions
• get_session_schema — view the scoring framework
• trigger_export — generate an Excel workbook

{
  "mcpServers": {
    "research-agent": {
      "command": "python",
      "args": ["mcp_server.py"],
      "cwd": "C:\\POD Company Research Agent"
    }
  }
}

Example: Scoring Companies with Claude

Once the MCP server is connected, you can ask Claude things like:

1. 1."Score TCS on all 7 dimensions based on the extracted data"
2. 2."What companies have we scraped but not scored yet?"
3. 3."Show me the data quality for session 1"
4. 4."Export the results to Excel"
5. 5."Compare the AI readiness scores across all analyzed companies"

Search Tips

Get better results from the Discover feature:

1. 1.Be specific: "Indian IT services companies AI ML consulting 500+ employees" works better than "tech companies"
2. 2.Run multiple searches with different angles (by industry, size, geography, specialization)
3. 3.The system filters out aggregator sites (Wikipedia, LinkedIn search pages, etc.)

Scraping Tips

Get the most out of web scraping:

1. 1.Companies with a website URL get the best data — the scraper visits multiple pages
2. 2.LinkedIn URLs add extra data (company size, headquarters, leadership)
3. 3.Scraping respects rate limits (3s between pages, 10s between companies)
4. 4.If a scrape fails, check the Activity page for the error and try again

Common Issues

1. 1.500 errors — check your backend terminal (Terminal 1) for the Python traceback
2. 2.Empty dashboard — make sure you have an active session selected in the top bar
3. 3.No companies showing — check that the session filter matches where your companies were imported
4. 4.Scrape stuck on "scraping" — the background task may have failed silently; check Activity log