A company research agent needs more than a plausible summary. It needs financial figures tied to a reporting period, company background from an identifiable source, and links you can follow to check its claims. We’ll build one in Langflow using Dataville’s MCP server. Using Walmart as our example, we’ll combine Wikipedia context with SEC EDGAR filings, generate a cited report, and inspect the tool responses to see where the numbers and dates came from. Want to run the example first? The companion repository includes the flow, agent instructions, and a sample report. The demo video builds the flow in this guide, from the Simple Agent template to the finished Walmart report. Before it ends, it opens Walmart’s 10-K on the SEC website to check the figures in the report.

Before you start

You’ll need:

Langflow. Install Langflow Desktop
A model provider (from OpenAI, Anthropic, OpenRouter, etc)
A Dataville API key for regular use. You can try the example without one, then get a key from the API Keys page when you’re ready.

Without a key, data lookups use the anonymous allowance of 50 per 24 hours per IP address, shared by everyone on that address. See Usage limits for the limits and Create an API key for authentication details.

Step 1: Connect Dataville to Langflow

Create a flow

Open Langflow and create a new flow. For this example, we’ll start from the Simple Agent template.

Configure agent

The template includes Web Search and URL tools. Remove them for this example so the agent gets its source material through Dataville. Keep Chat Input connected to the Agent, and the Agent’s response connected to Chat Output.

Choose your model and add its provider key. The run in this guide used Claude Opus 4.7 through OpenRouter. You can use another tool-calling model supported by Langflow, but the wording and tool choices may differ.

Add MCP Server

Open MCP in the left sidebar and click Add MCP Server.

Select Streamable HTTP/SSE and enter the connection details:

Field	Value
Name	`dataville`
Streamable HTTP/SSE URL	`https://api.dataville.com/mcp`
Header key	`Authorization`
Header value	`Bearer YOUR_API_KEY`

Replace YOUR_API_KEY with your Dataville key, or leave out the header to try anonymous access.

Click Add Server. The Connect an AI tool page is the setup reference if you need to check the configuration later.

Step 2: Give the agent access to Dataville tools

Import MCP Server

Drag the server from the MCP sidebar onto the canvas. Langflow adds an MCP Tools component for that server.

Enable Tool Mode

Select the component and turn on Tool Mode in its header menu. The component exposes a Toolset output, which you connect to the Tools input on the Agent. The connection makes the tools available to the agent.

Add Agent Instructions

Agent instructions help tell the model when to use them and what to return. For this example, we want Wikipedia background kept separate from SEC-reported financials, with a citation for each section.Paste these instructions into Agent Instructions:

You are a company research assistant. Use Dataville to build a sourced report on the company the user names.

You have two tools:
- list_dataville_sources lists the available sources.
- search_dataville searches a source by keyword.

For a company report:
1. If needed, list the available sources.
2. Search the wikipedia source for the company's background, history, headquarters, and markets. Use its full name to avoid ambiguous matches.
3. Search the edgar source for the latest 10-K. Include the form type in the keywords, for example "Walmart 10-K".
4. Search the edgar source separately for the latest 10-Q, for example "Walmart 10-Q".
5. Use recent_filings and insider_transactions from the returned metadata when available. These are limited lists, not a complete filing or transaction history.
6. Confirm each result belongs to the requested company. If a result is missing or belongs to another company, say so. Do not substitute another company's filing.

Return a concise report with a company snapshot, annual and interim financials, recent filings, insider activity, and Wikipedia context.

For financials, state the start and end of the reporting period. Do not label year-to-date figures as a single quarter's results. Keep Wikipedia context separate from SEC facts.

Cite the source name, relevant date, and a URL supplied in the results for each section. Do not invent citation links. If an individual filing link is unavailable, say so. Preserve Wikipedia attribution and usage terms from the result.

Do not invent numbers, dates, or filing details. Flag stale results and missing information. Do not infer an insider's motives from a reported transaction.

The 10-K response used in this example also contains recent filings and insider transactions, so the agent doesn’t need a Form 4 lookup. It still searches for the 10-Q to get the interim financials. The insider activity in the report comes from the insider_transactions field in the lookup’s metadata.

Step 3: Run the company research agent

Generate report

Open the Playground and send this prompt:

Build a company research report on Walmart.

The chat history shows the agent calling Dataville for the source material.

Check the tool calls

If you click on one of the agent tool calls, you can inspect the arguments used for the query. In this instance the agent searched for Walmart 10-K in the edgar data source.

To learn more about the JSON response, including the metadata each source returns, see Response format.

Check the report

The report includes a company snapshot, financials, recent filings, insider activity, and Wikipedia context. Each section cites the source, date, and URL from the result.

To validate the report, follow the citations provided by the agent and confirm the numbers found in the SEC filings match up with the information in the report.

Get the Langflow example project

The datavilleorg/langflow-company-research repository contains the flow, agent instructions, and sample output. Add the hosted MCP server, import the flow, choose your model, and run the company prompt. Clone the repo and run the flow.

Underwriting and onboarding

Market research

Sales and revenue operations

Build a Langflow Agent with SEC EDGAR and Wikipedia

Before you start

Step 1: Connect Dataville to Langflow

Step 2: Give the agent access to Dataville tools

Step 3: Run the company research agent

Get the Langflow example project

​Before you start

​Step 1: Connect Dataville to Langflow

​Step 2: Give the agent access to Dataville tools

​Step 3: Run the company research agent

​Get the Langflow example project

Before you start

Step 1: Connect Dataville to Langflow

Step 2: Give the agent access to Dataville tools

Step 3: Run the company research agent

Get the Langflow example project