deeplook
# DeepLook
LLMs hallucinate financial data. DeepLook gives them real numbers instead.
[](https://www.gnu.org/licenses/agpl-3.0)
[](https://github.com/OSOJDJD/deeplook)
## What happens when you ask "Research NVIDIA"
DeepLook provides structured context — real-time data from 8+ APIs combined with analytical instructions that makes better output. The result:
- **Accurate, up-to-date information** — not hallucinated numbers
- **Clear at a glance** — financials, peers, technicals, news in one view
- **Better AI output** — because good context drives good analysis
Works for financial research, business due diligence, or any use case where you need to understand a company fast.
---
## Under the hood
- **10+ sources, one call** — queries fan out in parallel across market data, filings, news, and more
- **Rules before reasoning** — data is verified before the LLM ever sees it, which is what keeps hallucination down
- **Eval-tested prompts** — changes get scored before they ship, not just shipped on vibes
- **Agent-ready** — available as both an MCP server and a REST API
---
## Get started
### Claude.ai
1. Go to **Settings → Connectors → Add MCP**
2. Paste: `https://mcp.deeplook.dev/mcp`
3. Start a new chat and ask: *"Research NVIDIA"*
### Other MCP clients (Cursor, VS Code, Windsurf, Claude Desktop)
```json
{
"mcpServers": {
"deeplook": {
"url": "https://mcp.deeplook.dev/mcp"
}
}
}
```
### Claude Code
```bash
claude mcp add --transport http deeplook https://mcp.deeplook.dev/mcp
```
### Self-host
```bash
git clone https://github.com/OSOJDJD/deeplook.git
cd deeplook
pip install -r requirements.txt
python -m deeplook.mcp_server
```
---
## What it covers
| Type | Examples |
|------|----------|
| Public stocks | NVIDIA, Apple, Tesla, TSMC |
| Crypto | Bitcoin, Solana, Ethereum |
| Private companies | Anthropic, Stripe, OpenAI |
| VC firms | a16z, Sequoia |
| Defunct | FTX, WeWork |
---
## Eval
Tested across 58 companies, scored on accuracy, hallucination, and usefulness:
| Metric | Score |
|--------|-------|
| Overall | 3.78 / 5.0 |
| Risk detection | 4.36 / 5.0 |
| Signal quality | 3.94 / 5.0 |
| Actionability | 3.38 / 5.0 |
*From an earlier pipeline version — re-run pending. Eval harness in [`/deeplook/eval`](deeplook/eval).*
---
## Extend DeepLook
DeepLook covers the basics. If you need data it doesn't have yet — a new market, a new data source, a new analysis rule — you can add it. See [CONTRIBUTING.md](CONTRIBUTING.md).
---
## Roadmap
- [ ] More query tools — news, peers, financials, calendar as standalone lookups
- [ ] Broader client support — Cursor, ChatGPT, VS Code, Windsurf, Claude Code
- [ ] Deeper context — more analytical conditions, entity-specific instructions
- [ ] Community contributions — new data sources, custom analysis rules
---
## License
[AGPL-3.0](LICENSE)
---
Built by [@OSOJDJD](https://github.com/OSOJDJD)
TDQS
Scored across 2 tools
The two tools have clearly distinct purposes: deeplook_lookup provides a quick snapshot for fast checks, while deeplook_research offers a comprehensive structured report from multiple data sources. There is no overlap or ambiguity between them.
Both tools follow a consistent 'deeplook_' prefix with descriptive suffixes (lookup and research), using snake_case uniformly. The naming pattern is predictable and well-structured.
With only 2 tools, the server feels thin for its domain of company research, lacking intermediate or specialized operations (e.g., updates, filtering, or historical analysis). While the tools cover basic needs, the scope is minimal.
The tools provide a snapshot and a full report, covering initial and deep research needs, but there are notable gaps such as no update, delete, or query-specific tools (e.g., by industry or date). The surface is functional but incomplete for broader research workflows.