EDGAR SEC Memory MCP
# FINSEC — Certified SEC Memory MCP Server
[](https://registry.modelcontextprotocol.io)
[](https://smithery.ai/servers/calera-labs/finsec)
[](https://glama.ai/mcp/servers/Calera-Labs/finsec-mcp)
[](https://glama.ai/mcp/servers/Calera-Labs/finsec-mcp)
[](LICENSE)
[](https://finsec.caleralabs.com)
[-purple.svg)](https://finsec.caleralabs.com)
Official **Model Context Protocol (MCP)** connector for **Calera Labs FINSEC**. Connects AI agents (Claude Desktop, Cursor, Zed, VS Code, and custom autonomous swarms) directly to **4.06M certified SEC EDGAR facts** powered by the **Calera Episteme-1 Engine**.
> **"Frontier Models Reason — The Calera Lattice Supplies Ground Truth."**
---
## ⚡ Key Guarantees
* **0.00% Statistical Hallucination:** Fact recall is deterministic and geometric ($A_4$ Pentatope lattice memory). No language model exists in the fact retrieval path.
* **2026-07-28 Stateless Protocol Core:** Fully compliant with the 2026-07-28 Stateless MCP Specification (SEP-2243, SEP-2575, SEP-2549) with sub-5ms zero-handshake direct tool calls and intelligent caching (`ttlMs: 86400000`).
* **Cryptographic Provenance:** Every returned financial metric includes exact SEC filing accessions, dates, concepts, and SHA-256 audit hashes.
* **Honest Refusal (`SAFE_REFUSAL`):** If a metric is not present in official SEC filings or is ambiguous, the engine returns an honest safe refusal rather than confabulating numbers.
* **Ultra-Low Latency & High Gross Margins:** Sub-25ms response serialization with low token footprint mode.
---
## 🚀 1-Click Quickstart Integrations
### 1. Cursor IDE
Add the following to your project's `.cursor/mcp.json` (or global Cursor Settings → Features → MCP):
```json
{
"mcpServers": {
"finsec": {
"url": "https://finsec.caleralabs.com/mcp",
"headers": {
"X-License-Key": "clabs_live_YOUR_KEY"
}
}
}
}
```
*(Get your free API key at [dashboard.caleralabs.com](https://dashboard.caleralabs.com))*
---
### 2. Claude Desktop
Add to your `claude_desktop_config.json`:
* **macOS:** `~/Library/Application Support/Claude/claude_desktop_config.json`
* **Windows:** `%APPDATA%\Claude\claude_desktop_config.json`
* **Linux:** `~/.config/Claude/claude_desktop_config.json`
```json
{
"mcpServers": {
"finsec": {
"url": "https://finsec.caleralabs.com/mcp",
"headers": {
"X-License-Key": "clabs_live_YOUR_KEY"
}
}
}
}
```
---
### 3. Smithery CLI (1-Click Terminal Command)
Install automatically across Claude or Cursor using Smithery ([smithery.ai/servers/calera-labs/finsec](https://smithery.ai/servers/calera-labs/finsec)):
```bash
# For Claude Desktop
npx -y @smithery/cli install financesec-mcp --client claude
# For Cursor
npx -y @smithery/cli install financesec-mcp --client cursor
```
---
### 4. Zed Editor
Add to `~/.config/zed/settings.json`:
```json
{
"experimental.context_servers": {
"finsec": {
"command": {
"url": "https://finsec.caleralabs.com/mcp",
"headers": {
"X-License-Key": "clabs_live_YOUR_KEY"
}
}
}
}
}
```
---
## 🛠️ Certified Tools Surface Area
The FINSEC MCP server exposes **7 certified tools**:
| Tool Name | Description |
|---|---|
| `query_financial_sec` | Natural-language query over 4.06M certified SEC EDGAR facts with provenance or honest `SAFE_REFUSAL`. |
| `valuation_inputs` | 12 Certified Valuation Packs (DCF inputs, FCF, Altman Z', Beneish M, DuPont 5-step, Piotroski F). |
| `query_sec_metric_exact` | Exact single-cell EDGAR metric recall for company + period + concept. |
| `compute_sec_cagr` | Exact compound annual growth rate (CAGR) calculation over verified SEC filing facts with zero confabulation. |
| `lattice_arith_evaluate` | Deterministic zero-error algebraic solver (`cagr`, `ratio`, `multiply`, `divide`, `yoy_series`, `product_chain`). |
| `vln_capabilities_overview` | Dynamic scope overview and capability boundaries. |
| `query_sec_sector_peers` | Multi-company sector & industry financial comparisons. |
---
## 💡 Example Queries for Your Agent
Once connected, simply ask your agent in plain English:
* *"What was Apple's FY2023 total revenue and gross margin in USD?"*
* *"Compare Microsoft and Alphabet Q3 2024 operating income."*
* *"Calculate Nvidia's 3-year revenue CAGR from FY2021 to FY2024."*
* *"Pull the DuPont 5-step analysis inputs for Tesla in 2023."*
---
## 🏎️ Low-Token Mode for High-QPS Agent Swarms
To eliminate context exhaustion in multi-turn reasoning agents, pass `"low_tokens": true` in your queries for a **71.7% token reduction**:
```json
{
"id": "q112",
"st": "OK",
"val": 383285000000,
"cur": "USD",
"concept": "us-gaap:RevenueFromContractWithCustomerExcludingAssessedTax",
"per": "FY2023",
"hash": "5ce23625"
}
```
---
## 🔑 Authentication & API Keys
Generate your free or enterprise API key on the **Universal Dashboard**:
👉 **[https://dashboard.caleralabs.com](https://dashboard.caleralabs.com)**
---
## 🛡️ Security & Provenance Policy
* **No Language Model in the Fact Path:** All returned numbers originate from XBRL/EDGAR filing tables verified against SEC accessions.
* **IP Cloud Boundary:** Substrate algorithms and $A_4$ Pentatope lattice weights operate inside Calera Labs secure cloud runtimes. Public clients receive only verified outputs and schema responses.
---
## 💬 Customer Support & Immediate Slack Pipeline
Have a question, encountered an edge case, or need a quota increase for your team?
We have a 3-tier Slack customer response pipeline wired directly into our engineering and executive teams:
### 1. Directly in Your AI Assistant (Claude, Cursor, Zed, Cline)
Use the built-in MCP tool `submit_support_ticket`:
```json
{
"title": "Need 50k req/min quota for quant backtesting cluster",
"description": "We are running systematic macro backtests across 500 tickers simultaneously.",
"inquiry_type": "quota_increase",
"priority": "high",
"contact_email": "quant-lead@firm.com"
}
```
### 2. Live Web Portal & Slack Escalation Center
Submit tickets, questions, or feature requests directly from the Universal Dashboard:
👉 **[https://dashboard.caleralabs.com/dashboard#customer_service](https://dashboard.caleralabs.com/dashboard#customer_service)**
### 3. Response SLAs
| Tier | Responder | SLA Guarantee |
| :--- | :--- | :--- |
| **Tier 1 (Normal)** | Customer Success Manager (`@manager_oncall`) | < 4 Hours |
| **Tier 2 (High / Quota)** | Operations Owner (`@owner_oncall`) | < 2 Hours |
| **Tier 3 (Critical / VIP)** | Founder & CEO Casey (`@casey`) | < 1 Hour |
---
## 🔗 Related Calera MCP Servers
* **[Calera Infinite Context (ICX) MCP](https://github.com/Calera-Labs/icx-mcp):** Persistent epistemic memory and sub-5ms associative recall on the Volumetric Lattice Network for AI agents ([Glama Hub](https://glama.ai/mcp/servers/Calera-Labs/icx-mcp) · [Smithery](https://smithery.ai/servers/calera-labs/icx-mcp)).
---
## 📄 License & Attribution
* Connector specification and client schemas are licensed under the [Apache-2.0 License](LICENSE).
* Powered by **Calera Episteme-1** / **Calera Labs** (DBA of Pyrotope Inc., Subsidiary of Calera Computing, Inc.).
TDQS
Scored across 7 tools
Several tools overlap heavily: compute_sec_cagr and lattice_arith_evaluate both perform deterministic CAGR/arithmetic over SEC facts, and query_financial_sec and query_sec_metric_exact both return certified SEC metric values with provenance or SAFE_REFUSAL. The detailed descriptions help slightly, but an agent cannot clearly determine which math or query tool to choose.
Tool names mix patterns: query_* tools are verb-first, but lattice_arith_evaluate is verb-last, and valuation_inputs / vln_capabilities_overview are noun phrases. The unexplained 'vln' abbreviation and inconsistent verb placement make the naming scheme unpredictable.
Seven tools is a reasonable size for an SEC/EDGAR valuation server and not excessive. However, the overlapping math and query tools mean the set is slightly inflated rather than each tool occupying a distinct slot.
Core workflows are covered: certified fact/metric queries, peer screening, valuation input packs, deterministic calculations, and capability discovery. Minor gaps such as missing dedicated company-profile or filing-metadata tools exist but agents can work around them.