Veroq
# veroq-mcp
**Verified AI. One function call.** Stop shipping hallucinations.
62 MCP tools. `shield()` any LLM output. Evidence chains on every claim. Finance as flagship vertical, with legal, research, compliance, and custom domains built in.
[veroq.ai](https://veroq.ai) | [API Reference](https://veroq.ai/docs) | [Pricing](https://veroq.ai/pricing)
## Install
```bash
npm install -g veroq-mcp
```
Get your API key at [veroq.ai/settings](https://veroq.ai/settings).
## Quick Start
Once installed, two tools handle most use cases:
```
veroq_ask "How is NVIDIA doing?" → price, technicals, earnings, sentiment, news
veroq_verify "Apple beat Q4 earnings by 20%" → SUPPORTED (92%), 5 sources, evidence chain
```
For multi-agent workflows, use the Verified Swarm:
```
veroq_run_verified_swarm { "query": "Analyze NVDA for a long position" }
→ 5 agents: planner → researcher → verifier → critic → synthesizer
→ Every step verified, escalation on high-stakes, full decision lineage
```
For domain-specific pipelines, use the Agent Runtime:
```
veroq_create_runtime { "vertical": "finance", "query": "Analyze NVDA" }
veroq_create_runtime { "vertical": "legal", "query": "GDPR data retention" }
```
## Why VeroQ?
| Capability | What you get |
|-----------|-------------|
| **Verified outputs** | Evidence chains, confidence breakdowns, source reliability scores on every response |
| **Enterprise safety** | Permission engine, decision lineage, human-in-the-loop escalation, full audit trails |
| **Multi-agent workflows** | Verified Swarm with 5-agent pipeline, budget control, parallel execution, caching |
| **Multi-domain runtime** | Finance (flagship), legal, research, compliance, custom verticals with domain-specific safety |
| **External MCP integration** | Securely proxy external APIs through VeroQ's permission engine, rate limiting, and audit |
| **Self-improvement** | Feedback loop flags low-confidence outputs, web search fallback fills data gaps |
| **Cost control** | 3 cost modes (cheap/balanced/premium), per-step budgets, credit transparency |
| **Observability** | Per-tool metrics, cache hit rates, escalation rates, feedback volume |
## IDE Setup
### Claude Desktop
Add to `~/Library/Application Support/Claude/claude_desktop_config.json`:
```json
{
"mcpServers": {
"veroq": {
"command": "veroq-mcp",
"env": { "VEROQ_API_KEY": "your-api-key" }
}
}
}
```
### Cursor
**Settings > MCP Servers > Add Server**: Name `veroq`, Command `veroq-mcp`, Env `VEROQ_API_KEY=your-api-key`.
## Tools (62)
### Hero Tools
| Tool | What it does | Cost |
|------|-------------|------|
| `veroq_ask` | Ask any financial question — routes to 41 intents automatically | 1-5cr |
| `veroq_verify` | Fact-check any claim with evidence chain + confidence breakdown | 3cr |
### High-Level Tools
| Tool | What it does | Cost |
|------|-------------|------|
| `veroq_analyze_ticker` | Complete ticker analysis (price, technicals, earnings, sentiment, news) | 3cr |
| `veroq_verify_market_claim` | Fact-check financial claims with evidence and source reliability | 3cr |
| `veroq_generate_trading_signal` | NLP screener + composite trade signal | 5cr |
| `veroq_comprehensive_intelligence` | Market briefing (indices, movers, trending, yields) | 3cr |
| `veroq_compare_tickers` | Side-by-side comparison with correlation matrix | 3cr |
| `veroq_tool_search` | Context-aware tool discovery with filtering | free |
### Swarm & Runtime
| Tool | What it does | Cost |
|------|-------------|------|
| `veroq_run_verified_swarm` | Multi-agent pipeline with verification at every step | 15-25cr |
| `veroq_create_runtime` | Domain-specific runtime (finance, legal, research, compliance) | 10-25cr |
| `veroq_process_feedback` | Submit or query the self-improvement feedback loop | free |
| `veroq_call_external_tool` | Securely call registered external MCP servers | 1-5cr |
### Market Data (16 tools)
`veroq_ticker_price` (free), `veroq_full`, `veroq_candles`, `veroq_technicals`, `veroq_earnings`, `veroq_market_movers`, `veroq_market_summary`, `veroq_economy`, `veroq_forex`, `veroq_commodities`, `veroq_sectors`, `veroq_portfolio_feed`, `veroq_screener`, `veroq_screener_presets`, `veroq_alerts`, `veroq_ticker_score`
### Crypto (4 tools)
`veroq_crypto`, `veroq_crypto_chart`, `veroq_defi`, `veroq_defi_protocol`
### Search & Intelligence (15 tools)
`veroq_search`, `veroq_feed`, `veroq_brief`, `veroq_extract`, `veroq_entities`, `veroq_trending`, `veroq_compare`, `veroq_research`, `veroq_timeline`, `veroq_forecast`, `veroq_contradictions`, `veroq_events`, `veroq_diff`, `veroq_web_search`, `veroq_crawl`
### Fundamentals (5 tools)
`veroq_insider`, `veroq_filings`, `veroq_analysts`, `veroq_congress`, `veroq_institutions`
### Other (7 tools)
`veroq_ticker_news`, `veroq_ticker_analysis`, `veroq_search_suggest`, `veroq_social_sentiment`, `veroq_social_trending`, `veroq_ipo_calendar`, `veroq_run_agent`
### Reports (2 tools)
`veroq_generate_report`, `veroq_get_report`
## Programmatic Usage
VeroQ is also available as a TypeScript/Python library:
```typescript
import { createRuntime, registerExternalMcpServer } from "veroq-mcp";
// Finance runtime with external market data
const runtime = createRuntime({
vertical: "finance",
enterpriseId: "acme-capital",
costMode: "balanced",
enableSelfImprovement: true,
externalServers: [{
serverId: "bloomberg",
serverUrl: "https://api.bloomberg.com",
auth: { type: "bearer", credential: process.env.BLOOMBERG_TOKEN },
allowedTools: ["get_security", "get_analytics"],
trustLevel: "read-only",
}],
});
const result = await runtime.run("Analyze NVDA for a long position");
console.log(result.synthesis.summary);
console.log(result.budget); // { spent: 12, remaining: 38 }
console.log(result.verificationSummary); // { avgConfidence: 82, flaggedSteps: 0 }
```
### Verify Any LLM Output
```typescript
// Extract and verify every claim in arbitrary LLM text
const verified = await client.verifyOutput("NVIDIA's Q4 revenue exceeded $22B and margins expanded to 75%");
console.log(verified.claims); // each claim with verdict, confidence, correction
console.log(verified.overall_confidence);
```
### Agent Memory
```typescript
// Store context tied to an agent
await client.memoryStore({ agent_id: "my-bot", key: "nvda-thesis", value: "bullish on AI capex" });
// Recall relevant context for a query
const context = await client.memoryRecall({ agent_id: "my-bot", query: "NVDA outlook" });
// List all stored memories
const memories = await client.memoryList({ agent_id: "my-bot" });
```
### Real-Time Verification Stream
```typescript
// Persistent SSE — pushes ticker_status, signal_change, confidence_change, claim_update
const stream = client.watch({ tickers: ["NVDA", "AAPL"], agent_id: "my-bot" });
stream.on("signal_change", (event) => console.log(event));
```
## Enterprise
Enterprise customers get:
- **Permission engine** — allow/deny/review rules with wildcard patterns
- **Decision lineage** — full rule evaluation trace for every tool call
- **Escalation** — high-stakes queries trigger human-in-the-loop review
- **Audit trails** — every decision logged with session tracking
- **External MCP** — securely proxy external APIs with trust levels and rate limits
- **Feedback loop** — self-improvement with web search fallback
- **Observability** — admin dashboard with verification stats, safety triggers, credit consumption
Contact [enterprise@veroq.ai](mailto:enterprise@veroq.ai) or visit [veroq.ai/pricing](https://veroq.ai/pricing).
## Environment Variables
| Variable | Required | Description |
|----------|----------|-------------|
| `VEROQ_API_KEY` | Yes | Your VEROQ API key |
| `VEROQ_BASE_URL` | No | API base URL (default: `https://api.veroq.ai`) |
`POLARIS_API_KEY` and `POLARIS_BASE_URL` accepted for backward compatibility.
## TradingAgents-Pro
[TradingAgents-Pro](https://github.com/Veroq-api/TradingAgents-Pro) is a demo showcasing what VeroQ can do — 15 agents running on the verified intelligence layer. **VeroQ is the product. TradingAgents-Pro is the showcase.**
## License
MIT
TDQS
Scored across 78 tools
Many tools have overlapping purposes, e.g., multiple ticker analysis tools (veroq_ticker_price, veroq_ticker_analysis, veroq_ticker_score, veroq_full, veroq_fast_snapshot) and multiple search tools (veroq_search, veroq_web_search, veroq_ask). Descriptions help but ambiguity remains.
Most tools follow a consistent 'veroq_verb_noun' pattern (e.g., veroq_get_report, veroq_run_agent). There are minor deviations like 'veroq_agent_packs' and 'veroq_fast_heatmap' but overall predictable.
78 tools is excessive for an MCP server. Many are niche (travel, GitHub, Hacker News) and could be consolidated. The server would benefit from reducing tangential features.
The tool set covers a wide range of financial data: prices, fundamentals, technicals, sentiment, news, insider, filings, macro, and more. Fact-checking and research tools add depth. Minor gaps like lack of portfolio management are acceptable.