For agents
Everything an AI agent needs to use Clous — a paste-in system prompt, MCP install, function-calling tool definitions, and the response envelope.
For agents
Clous is built to be driven by AI agents. Drop the block below into your agent and it knows the base URL, auth, the one response envelope, and the endpoints to reach for.
Paste this into your agent
Add to CLAUDE.md, .cursorrules, or your system prompt:
# Clous — agent-native SEC/EDGAR API
BASE https://api.clous.ai
AUTH Authorization: Bearer ${CLOUS_API_KEY}
DOCS https://docs.clous.ai/llms.txt (full corpus: /llms-full.txt)
Every response uses ONE envelope:
{ data[], page{ limit, next_cursor, has_more }, as_of, source, query_echo, warnings }
Paginate by passing page.next_cursor back as ?cursor=. No offsets, no result ceiling.
Endpoints you'll use most:
GET /v1/entities?ticker=AAPL resolve a company -> CIK / entity_id
GET /v1/filings?cik=&form_type=&q= EDGAR filing index (all forms)
GET /v1/full-text?q="going concern" full-text search of filing bodies (2001+)
GET /v1/financials/{cik} XBRL company facts
GET /v1/insider?ticker=AAPL Form 3/4/5 insider transactions
GET /v1/raises?min_amount=5000000 Form D private placements
GET /v1/cyber-incidents 8-K Item 1.05 cybersecurity disclosures
GET /v1/events normalized business-change event feed
POST /v1/monitors watch a ticker/form/event -> signed webhooks
Always read query_echo to confirm the params Clous applied.
Get a key (100 free credits): https://clous.aiModel Context Protocol
Connect once; your agent gets typed tools. Use the hosted server (zero install) or
the open-source @clousai/mcp package locally.
{
"mcpServers": {
"clous": {
"type": "http",
"url": "https://mcp.clous.ai",
"headers": { "Authorization": "Bearer clous_live_…" }
}
}
}{
"mcpServers": {
"clous": {
"command": "npx",
"args": ["-y", "@clousai/mcp"],
"env": { "CLOUS_API_KEY": "clous_live_…" }
}
}
}Works with any MCP client — Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Zed. See the MCP setup page.
Function-calling tool definition
Prefer raw function calling? Register Clous endpoints as tools. The full set is in the OpenAPI 3.1 spec.
{
"type": "function",
"function": {
"name": "clous_full_text_search",
"description": "Full-text search across the body of every EDGAR filing since 2001.",
"parameters": {
"type": "object",
"required": ["q"],
"properties": {
"q": { "type": "string", "description": "Keyword or \"exact phrase\"." },
"forms": { "type": "string", "description": "Comma-separated forms, e.g. 8-K,10-K." },
"limit": { "type": "integer", "description": "1-100, default 25." }
}
}
}
}The response envelope
Every endpoint returns the same shape, so an agent learns it once:
{
"data": [ /* records */ ],
"page": { "limit": 25, "next_cursor": "…", "has_more": true },
"as_of": "2026-06-13T08:36:42Z",
"source": "form_d",
"query_echo": { "min_amount": 5000000, "limit": 25 },
"warnings": []
}Machine-readable indexes: /llms.txt (endpoint list)
and /llms-full.txt (full corpus). Point your
agent at either to ingest the whole API.
Real-time material-event alerts
Create a monitor and a signed webhook, and get pushed the moment a matching SEC filing or event lands — no polling.
MCP
The Model Context Protocol server for Clous — run it locally with npx @clousai/mcp or connect to the hosted server at mcp.clous.ai, so AI agents can call Clous tools directly.