This is a submission for the Notion MCP Challenge
What I Built
A system that turns a single prompt into a structured corporate intelligence workspace — using two MCP servers to bridge government data sources and Notion.
The problem: Investigating a company's compliance posture means manually checking state business registries, SEC filings, federal contracts, and lobbying disclosures. Each source has a different interface, different data formats, and no cross-referencing. A thorough review of one company across these sources takes a human analyst hours. Three companies? Days.
The solution: Filed MCP Server provides 6 tools covering 9 US state business registries and 3 federal data sources. Notion MCP Server provides the structured workspace. One natural language prompt triggers a multi-source investigation and populates a Notion hub with cross-referenced findings.
I demonstrated this by investigating three defense contractors — Lockheed Martin, Anduril Industries, and Palantir Technologies. In 37 autonomous AI turns ($0.91 API cost), the system:
- Searched all three companies across 9 state databases, SEC EDGAR, USASpending, and Senate lobbying records
- Created and populated two Notion databases (Companies + Intelligence Findings)
- Surfaced 13 actionable intelligence findings — including 3 Critical compliance risks the AI identified by cross-referencing registration gaps against known business locations
The standout finding: Anduril Industries operated as an unregistered foreign corporation in 5 US states for 4-5 years while accumulating $784M in federal contracts. Their registration sprint in August 2024 — filing in 5 states within 13 days — suggests a retroactive compliance fix. A human analyst would need days to surface this pattern. The AI found it in minutes.
Video Demo
Live Notion workspace: Filed Intelligence Hub
What the AI Found (Real Data)
| Company | States | Entities | Federal Contracts | Lobbying | Risk |
|---|---|---|---|---|---|
| Lockheed Martin | 8/9 | 71 | $5.04B | $8.98M | Medium |
| Anduril Industries | 6/9 | 6 | $784.8M | $4.21M | High |
| Palantir Technologies | 3/9 | 13 | $553.2M | $4.96M | High |
Critical findings the AI surfaced on its own:
- Lockheed Martin — No Colorado registration despite LM Space HQ in Littleton, CO. Revoked entity in DC (Federal Healthcare Inc.)
- Anduril Industries — No Florida registration despite $458M+ in CBP border contracts. All 5 non-DC registrations filed within a 13-day window in Aug-Sep 2024
- Palantir Technologies — No Colorado registration 5+ years after HQ move to Denver (confirmed via SEC 8-K). SEC 10-K lists Florida as a business location, but no FL registration exists
Show us the code
GitHub: github.com/mgrantley/filed-mcp-server
Install:
npx filed-mcp-server
MCP Configuration:
{
"mcpServers": {
"filed": {
"command": "npx",
"args": ["-y", "filed-mcp-server"],
"env": {
"FILED_API_KEY": "your_key_here"
}
},
"notion": {
"command": "npx",
"args": ["-y", "@notionhq/notion-mcp-server"],
"env": {
"OPENAPI_MCP_HEADERS": "{\"Authorization\": \"Bearer YOUR_NOTION_TOKEN\", \"Notion-Version\": \"2022-06-28\"}"
}
}
}
}
Get a free API key at filed.dev/developers — 100 lookups/month on the free tier.
Filed MCP Server — 6 Tools
| Tool | What it does |
|---|---|
filed_search_entities |
Search business entities by name across 9 US states |
filed_get_entity |
Full entity details — officers, registered agent, filing history |
filed_company_intel |
Orchestrates all sources into one unified intelligence report |
filed_search_sec |
Search SEC EDGAR filings (10-K, 10-Q, 8-K, S-1) |
filed_search_contracts |
Search federal contracts from USASpending.gov |
filed_search_lobbying |
Search Senate lobbying disclosures (LDA database) |
The filed_company_intel tool is the key differentiator. One call searches all 9 states, pulls SEC filings, federal contracts, and lobbying data, then returns a structured report with entity counts, name variations, officer data, and total dollar values.
Architecture:
AI Agent (Claude Code / Cursor / Claude Desktop)
├── Filed MCP Server ← 6 tools, ~500 lines TypeScript
│ └── Filed.dev API
│ ├── State registries (AK, CO, CT, DC, FL, IA, NY, OR, PA)
│ ├── SEC EDGAR
│ ├── USASpending.gov
│ └── Senate LDA
└── Notion MCP Server ← structures results into workspace
└── Notion API
├── Companies (profiles with risk levels)
└── Intelligence Findings (cross-referenced analysis)
How I Used Notion MCP
Notion MCP is the persistence and analysis layer. Without it, Filed API responses are ephemeral chat messages. With it, the AI builds a structured, queryable workspace that persists.
Database creation: The AI used Notion MCP to create two databases with typed properties — numbers for contract values, selects for risk levels and severity ratings, rich text for detailed findings. The schema maps directly to the intelligence output.
Data population: After each filed_company_intel call, the AI populated structured rows with real data — entity counts, dollar-formatted contract values, lobbying totals, and multi-line key findings summaries.
Cross-referencing (the key insight): Because data persists in Notion, the AI could analyze patterns across companies after all three were researched:
- Companies missing state registrations where they clearly operate (HQ locations, contract sites)
- Registration timing patterns (Anduril's 13-day sprint across 5 states)
- Lobbying intensity ratios compared across peers (Palantir at 0.90% vs Lockheed at 0.18%)
- Entities with non-Active status (Revoked, Withdrawn, Non-Compliant) flagged as compliance risks
Intelligence Findings database: Each finding includes severity (Critical/Warning/Info), detailed description, recommended action, and specific data sources cited. This isn't a data dump — it's structured analysis a compliance officer could act on immediately.
Why This Pairing Works
Filed MCP handles the hard part: normalizing data from 9 different state business registries (each with its own format), SEC EDGAR, USASpending, and the Senate lobbying database into a consistent API.
Notion MCP handles the output part: turning structured data into a navigable, filterable workspace with typed properties and relational structure.
Neither server alone is remarkable. Together, they turn "research this company for due diligence" into a one-prompt operation that produces actionable intelligence with cited sources.
37 turns. 13 findings. $0.91. Zero human intervention.
Built by @grantleydev. Filed is a free US business entity search API.
Top comments (0)