# Build a Claude and Quiver Market Research Alert System Connect live alternative market data, run higher signal screens, and design Telegram alerts for new insider activity. ## What you will achieve You will create a Claude based workflow for researching insider trades, congressional disclosures, institutional holdings, contracts, and lobbying data, with a practical blueprint for automated alerts. ## Who this is for Investors, traders, and equity researchers who want to query structured market filings with natural language. **Difficulty:** Intermediate ## Short tutorial Connect live alternative market data, run higher signal screens, and design Telegram alerts for new insider activity. ### Guide: Build an AI Market Research System with Claude and Quiver **Time:** 25 to 40 minutes This tutorial shows you how to connect Quiver Quant data to Claude, interrogate filings with copy ready prompts, and turn useful screens into Telegram alerts. Start with the research workflow, validate the results, then add automation only after your alert criteria produce useful signals. The source quotes a Quiver hobbyist plan at approximately $30 per month. A paid plan may be required for proper API access. Claude access and an optional always running server may add separate costs. This workflow supports research, not automatic investment decisions. Insider purchases, sales, congressional disclosures, and institutional positions all require context. ## 1. Prepare the data connection You will need: * Access to Claude, preferably Claude Code for the complete workflow * A Quiver Quant account * A Quiver API key if required by your plan * Telegram if you want phone alerts * A computer or server that can keep the alert process running Quiver organizes public but difficult to use records into queryable datasets. The datasets emphasized here are: | Dataset | Best use | Important limitation | |---|---|---| | Corporate insider filings | Recent executive buying and selling | Transactions may reflect compensation, vesting, options, or planned sales | | Congressional trades | Political context and historical research | Disclosures may arrive weeks after the trade | | Institutional holdings | Tracking major holders and fund positioning | Reported holdings may not represent current positioning | | Federal contracts | Measuring government exposure | Awards need company and contract context | | Lobbying disclosures | Finding programs and issues a company is pursuing | Lobbying activity does not guarantee future revenue | ### Connect through Claude Code 1. Create your Quiver account. 2. Open the Quiver dashboard and create an API key. 3. Open Claude Code in the folder where you want the research or alert project stored. 4. Ask Claude Code to establish the Quiver connection using the available API or MCP configuration. 5. Confirm the connection with a small query before running a market wide screen. Use this setup request: ```text Help me connect this Claude Code project to Quiver Quant using my available API credentials or MCP configuration. First, inspect the available integration method and explain what configuration is required. Then create the local project structure needed to query Quiver data. Do not claim the connection works until you have run a small test query. For the test, request the most recent available insider transactions and show the filing date, transaction date, ticker, insider name, title, transaction type, share count, and reported value when those fields are available. If a required endpoint, package, permission, or field is unavailable, stop and tell me exactly what is missing. ``` Keep credentials out of prompts or files that will be shared publicly. Follow the credential storage method recommended by the integration you are using. ### Optional chat connector Claude settings also support connectors. Open **Settings**, select **Connectors**, and add the Quiver connector if it is available to your account. This is useful for interactive chat, but the source found Claude Code more reliable for building the automated workflow. ## 2. Verify freshness and provenance Do not begin with a complex screen. First establish which dates and fields Claude can actually retrieve. ```text Using only the connected Quiver data, show me the 20 most recently filed corporate insider transactions. For every row, include: 1. Company and ticker 2. Insider name and title 3. Purchase, sale, option exercise, or other transaction classification 4. Transaction date 5. Filing date 6. Share count 7. Reported dollar value 8. The Quiver dataset or record used Sort by filing date, newest first. Clearly label missing fields. Do not fill gaps with assumptions or general web results. ``` Check that the output distinguishes the transaction date from the filing date. This matters because a recently filed disclosure may describe an older transaction. Corporate insider filings are generally more suitable for timely alerts. Congressional disclosures are better treated as research context because the source found a substantial delay between trade and filing dates. ## 3. Research insider positioning for one company Start with a specific ticker before scanning the whole market. ```text Using Quiver insider trading data, analyze whether insiders at [TICKER] are buying or selling. Use the most recent available records and separate: 1. Open market purchases 2. Open market sales 3. Option exercises 4. Grants, vesting, or compensation related transactions 5. Other transaction types Show the transaction date, filing date, insider name, title, shares, price, and reported dollar value when available. Then summarize: 1. The number of distinct buyers and sellers 2. Total reported purchase and sale values 3. Whether activity appears concentrated among senior executives 4. Whether several records were filed on the same date or have similar values 5. Any reason the raw totals could be misleading Do not interpret insider activity as a recommendation to buy or sell. ``` A sale is not automatically bearish. An insider may be diversifying, realizing gains from an early holding, exercising options, or following a compensation plan. A purchase can also require context, but open market buying removes some of the ambiguity associated with awards and option exercises. ## 4. Screen for clusters of open market purchases The central market wide screen looks for companies with several distinct insiders buying shares. ```text Using Quiver insider trading data, find every company where three or more distinct insiders made open market purchases during the last 60 days. Requirements: 1. Include only open market purchases. 2. Exclude option exercises, grants, awards, vesting transactions, and sales. 3. Group results by company and ticker. 4. Rank first by the number of distinct buyers, then by total reported dollar value. 5. Show every buyer's name, title, transaction date, filing date, shares purchased, price, and reported value when available. 6. Flag transactions that appear to be part of a coordinated batch or compensation process. 7. Explain each batch flag using observable evidence, such as identical filing dates, similar transaction values, repeated titles, or tightly grouped transaction dates. 8. Separate potentially independent clusters from likely administrative batches. 9. State the latest filing date included in the analysis. Do not describe a cluster as bullish without discussing the transaction sizes and batch risk. ``` ### Detect false clusters A large buyer count can be misleading. The source gives an example in which 31 buyers initially looked extraordinary, but the purchases were approximately $40,000 each, occurred at vice president level, and were filed together. That pattern could indicate a compensation process rather than 31 independent decisions. For any promising result, run this drill down: ```text Stress test the insider purchase cluster for [TICKER]. Determine whether the purchases look like independent open market decisions or a coordinated administrative batch. Compare transaction dates, filing dates, reported values, share counts, insider titles, and transaction codes. Present: 1. Evidence supporting an independent cluster 2. Evidence supporting a batch or compensation explanation 3. The largest and smallest purchase 4. Purchases by senior officers or directors 5. Any outlier transaction that differs materially from the rest 6. A conclusion labeled strong cluster, mixed evidence, or likely batch Base the conclusion only on available records and state any missing information. ``` ## 5. Use congressional activity as company context A recent filing is not necessarily a recent trade. Always display both dates. ```text Using Quiver congressional trading data, show the most recently filed congressional trades. For each record include: 1. Member name 2. Chamber and party, if available 3. Ticker or asset 4. Purchase or sale classification 5. Reported value range 6. Trade date 7. Filing date 8. Days between the trade and filing Sort by filing date, newest first. Add a warning to any record where the filing delay makes it unsuitable as a real time signal. ``` For deeper company research, combine congressional trades with contracts and lobbying disclosures: ```text Give me the full political exposure profile for [COMPANY OR TICKER] using Quiver data. Analyze: 1. Congressional purchases and sales 2. Federal contract awards 3. Lobbying spending and disclosures For congressional activity, show trade dates and filing dates separately. For contracts, summarize reported award values, agencies, and timing when available. For lobbying, show spending and the specific issue text from individual disclosures when available. Finish with: 1. Areas where political or government exposure appears to be increasing 2. Areas where the evidence is mixed 3. Important reporting delays or missing fields 4. Questions that require further company research Do not treat lobbying, contracts, or congressional trades as proof of future revenue or stock performance. ``` The lobbying issue text can be especially useful because it may reveal the programs or policy areas a company is pursuing before those efforts appear as revenue. ## 6. Combine the outside and inside views The political profile is the outside view. Insider behavior and institutional ownership provide an inside ownership view. ```text Give me a complete ownership and sentiment profile for [TICKER] using connected Quiver data. Include: 1. Insider purchases and sales by quarter 2. Net reported insider activity by quarter 3. Institutional ownership 4. Top institutional holders 5. Material reported increases or reductions by major holders 6. Any disagreement between insider behavior and institutional positioning Distinguish confirmed data from interpretation. If insiders are selling while institutions are split, describe that as disagreement rather than clear directional momentum. ``` This approach prevents a single dataset from dominating the thesis. Consecutive insider selling, for example, may look more significant when major institutions are also reducing exposure. It may look less conclusive when institutional positioning is mixed. ## 7. Search for early, uncrowded activity Use this screen to find tickers where participation remains limited but activity is increasing: ```text Using Quiver congressional and corporate insider data from the last 90 days, identify tickers traded by fewer than five distinct people where activity appears to be increasing. Compare those names with tickers that already appear crowded. You must define increasing activity using measurable fields, such as: 1. More distinct participants in the recent period 2. Higher transaction frequency 3. Higher reported dollar value 4. A shift from selling to buying 5. Shorter gaps between relevant transactions For each result, show the underlying records and explain which measurements caused it to qualify. Rank by signal quality, but do not equate the ranking with expected returns. ``` Follow with: ```text From the results above, rank the stocks that deserve the most additional research based only on the congressional and insider patterns. For each stock provide: 1. The observed pattern 2. Why it may be notable 3. The strongest alternative explanation 4. Filing delay concerns 5. The next fundamental question to investigate Label the ranking as a research priority list, not a buy list. ``` ## 8. Define useful alert rules Start with a narrow alert instead of forwarding every filing. A practical rule can monitor open market purchases while excluding option exercises and likely batches. Copy and customize this specification: ```text Create an alert specification for Quiver corporate insider filings. Alert when all of the following are true: 1. The transaction is an open market purchase. 2. The ticker is [ANY TICKER OR WATCHLIST]. 3. The reported purchase value is at least [MINIMUM VALUE]. 4. The insider title matches [TITLE FILTER OR ANY]. 5. The filing has not already generated an alert. Increase priority when: 1. Two or more distinct insiders buy the same ticker within [NUMBER] days. 2. A chief executive, chief financial officer, director, or other selected senior role participates. 3. The purchase is materially larger than the insider's other recent purchases. Reduce priority or flag for review when: 1. Several transactions have nearly identical values. 2. Many records share the same filing date. 3. The activity appears compensation related. 4. Required transaction fields are missing. Each alert must include the ticker, company, insider, title, transaction date, filing date, shares, price, reported value, cluster count, and batch warning. ``` ## 9. Build the Telegram alert bot The source does not provide exact API endpoints or finished bot code, so let Claude Code inspect your connected environment and generate the implementation without inventing unavailable fields. ### Create the Telegram bot 1. Open Telegram and start a conversation with BotFather. 2. Create a bot and obtain its token. 3. Make the token available to your Claude Code project using an appropriate secret or environment configuration. 4. Establish the destination chat needed to receive messages. Then give Claude Code this build prompt: ```text Build a Telegram alert service for the connected Quiver data source. The service must: 1. Check for newly available corporate insider filings. 2. Apply the alert specification below. 3. Send matching records to my Telegram bot. 4. Store a stable filing or transaction identifier so the same record is never sent twice. 5. Include transaction date and filing date in every alert. 6. Separate open market purchases from option exercises, grants, vesting, and sales. 7. Add a likely batch warning when records have identical filing dates, similar values, or other observable batch characteristics. 8. Log successful checks, errors, and sent alerts. 9. Recover cleanly after temporary API or network errors. 10. Provide a test mode that uses a known historical record without sending duplicate live alerts. Before writing the implementation, inspect the available Quiver connection and list the actual fields and methods you can use. Do not invent an endpoint, field, or streaming capability. If the data source does not support event delivery, implement the most appropriate periodic check and explain the resulting delay. Use configuration values for: QUIVER credentials TELEGRAM bot token TELEGRAM destination chat Check frequency Watchlist Minimum transaction value Included insider titles Cluster window Alert specification: [PASTE YOUR COMPLETED ALERT SPECIFICATION HERE] After implementation, provide exact instructions to configure, test, start, stop, and monitor the service. ``` A Telegram notification format could look like this: ```text INSIDER PURCHASE ALERT Ticker: [TICKER] Company: [COMPANY] Insider: [NAME] Title: [TITLE] Transaction date: [DATE] Filing date: [DATE] Shares: [COUNT] Price: [PRICE] Reported value: [VALUE] Cluster: [COUNT] distinct buyers within [WINDOW] Batch warning: [NONE OR EXPLANATION] Record status: [NEW] ``` ## 10. Keep the service running A process running only on your laptop stops when the computer sleeps, shuts down, or loses its connection. For continuous alerts, run it on one of these: * A desktop computer that remains on * A dedicated device running continuously * An always running virtual server The source estimates that a basic virtual server may cost around $10 per month because the task mainly queries a dataset and sends notifications. Actual requirements and pricing depend on the generated implementation. Before relying on alerts, confirm: * A historical test record can reach Telegram. * Repeated checks do not resend the same record. * Purchases and sales are classified correctly. * Option exercises and awards are excluded when requested. * Filing dates and transaction dates are both visible. * Missing data produces a warning rather than an invented value. * Batch patterns are flagged for human review. * The service restarts successfully after an interruption. ## 11. Save a reusable prompt library Store the prompts you use successfully in a local folder, notes application, or phone. Organize them by purpose: ```text /prompts connection-test.md single-ticker-insiders.md purchase-cluster-screen.md batch-stress-test.md congressional-recent-filings.md political-exposure.md ownership-profile.md early-activity-screen.md alert-specification.md ``` You can also ask Claude to tailor additional queries to your research style: ```text I use Quiver data for [NEWS TRADING, FUNDAMENTAL RESEARCH, OR LONG TERM INVESTING]. My usual research universe is [MARKET, SECTOR, OR WATCHLIST]. My holding period is [PERIOD]. The risks I care about most are [RISKS]. Create 10 Quiver research prompts that fit this process. Each prompt must identify the required dataset, distinguish transaction dates from filing dates, request the underlying records, and include at least one alternative explanation for the observed pattern. ``` The strongest workflow is simple: screen broadly, inspect the underlying transactions, flag administrative batches, add company fundamentals and political context, then place only well understood names on a watchlist. Alerts should bring new information to your attention, not replace verification or judgment.