OpenCapital MCP Server

Every US public company, sourced from SEC filings — now available to your AI assistant.

OpenCapital is a hosted Model Context Protocol (MCP) server that gives Claude, ChatGPT, Cursor, Codex, and any other MCP-capable client read-only access to what every US public company actually reports — fundamentals, segment breakdowns, peer comparisons, and earnings for thousands of US-listed equities (common shares and ADRs), all derived directly from SEC filings.

You don't run anything. The server is hosted on OpenCapital's infrastructure. You add one URL to your client, sign in once, and start asking.

https://api.opencapital.sh/mcp

• Transport: Streamable HTTP (remote MCP server)
• Auth: OAuth 2.0 — sign in with Google, grant read-only access
• Free tier: 100 queries/day, no credit card. Paid plans remove the cap.
• Data: Fundamentals, segments, and earnings derived from US SEC filings (10-K, 10-Q, 8-K), read live

The data is free, the workflow is paid. Every metric, segment, quarter, and filing reference is free to query. OpenCapital is an independent project — not affiliated with the SEC.

---

Quick connect

Pick your client below, or see the full step-by-step guides at opencapital.sh/connect-ai.

Claude (claude.ai / Claude Desktop)

1. Customize → Connectors → Add custom connector
2. Name it OpenCapital, paste https://api.opencapital.sh/mcp
3. Click Connect and authorize read-only access

Full guide: opencapital.sh/connect-ai/claude

Claude Code

claude mcp add --transport http opencapital https://api.opencapital.sh/mcp

Add --scope user to make it available in every project. Then run /mcp inside a session, select opencapital, and choose Authenticate.

Full guide: opencapital.sh/connect-ai/claude-code

ChatGPT (Plus / Team / Enterprise)

1. Settings → Apps → Advanced settings → enable Developer mode
2. Create app, name it OpenCapital, paste the URL, select OAuth
3. Authorize read-only access

Full guide: opencapital.sh/connect-ai/chatgpt

Cursor

Add to ~/.cursor/mcp.json (global) or .cursor/mcp.json (project):

{
  "mcpServers": {
    "opencapital": {
      "url": "https://api.opencapital.sh/mcp"
    }
  }
}

Then enable the server under Cursor Settings → Tools & MCP → Connect.

Full guide: opencapital.sh/connect-ai/cursor

Codex / Codex CLI

codex mcp add opencapital --url https://api.opencapital.sh/mcp
codex mcp login opencapital

Or add to ~/.codex/config.toml:

[mcp_servers.opencapital]
url = "https://api.opencapital.sh/mcp"

Full guides: codex · codex-cli

Any other MCP client

Add a remote / custom MCP server over the Streamable HTTP transport, pointing at https://api.opencapital.sh/mcp. Most clients also accept the mcpServers JSON shape shown in the Cursor section above. The first call opens your browser to sign in and approve read-only access.

Full guide: opencapital.sh/connect-ai/other

---

What you can ask

The server ships with 14 read-only tools. You rarely call them by name — just ask in natural language and your client picks the right one.

Tool	What it does
stock_screener	Find and filter companies (sector, market cap, P/E, growth, FCF…). One row per company.
stock_screener_filter_values	List valid values for a stock filter field (e.g. count of stocks per sector).
metric_screener	Search financial metrics across companies. One row per (company × metric), with latest value, YoY/QoQ, and TTM change.
metric_screener_filter_values	List valid values for a metric filter field.
metric_detail	Definition, plain-language meaning, signal interpretation, parent/standard metric, and similar metrics across peers.
metric_time_series	Full quarterly / annual / TTM history for one or more metrics on a single ticker.
peer_time_series	One metric across up to 10 tickers — head-to-head peer comparisons.
segment_time_series	Segment disaggregation (product, business, geography) for a line item, with history per segment.
business_overview	What a company does, from its latest 10-K Item 1: segments, value chain, partnerships, named competitors.
business_competitors	Ranked peer list — competitors named in filings, reverse mentions, and embedding-similar businesses.
filing_text_search	Semantic + keyword search over the narrative prose of SEC filings (MD&A, risk factors, footnotes) — ranked passages with EDGAR links and cross-referenced metrics.
portfolio_holdings	Your linked brokerage holdings, aggregated by ticker with weights and P&L. (requires sign-in + linked account)
get_watchlist	Your watchlist with company metadata and nearest earnings event. (requires sign-in)
update_watchlist	Add or remove tickers from your watchlist. (requires sign-in)

Example prompts

• "What's Uber's Mobility take rate over the last 8 quarters?"
• "How does Airbnb's ADR compare across regions?"
• "Show me DoorDash Marketplace GOV every quarter since IPO."
• "Compare cloud revenue across AWS, Azure, GCP, and Oracle."
• "Find me 10 large-caps growing revenue over 20% with positive free cash flow and a P/E under 30."
• "Before Nvidia reports, summarize what their suppliers, hyperscaler customers, and competitors said in their last filings."

More in examples/prompts.md.

---

Read-only by design

The OpenCapital MCP server only exposes read-only stock research data — fundamentals, segments, and earnings derived from public SEC filings. Your assistant cannot place trades or modify any account state. The watchlist tools are the only write surface, and they only edit your own OpenCapital watchlist.

Privacy & terms

OAuth grants your AI client read-only access to OpenCapital's public stock research index (and, if you choose, your own watchlist and linked-brokerage holdings). See the Terms of Service and Privacy Policy.

License

The configuration snippets and documentation in this repository are MIT-licensed.

The OpenCapital MCP service they connect to is a hosted service governed by its Terms of Service — your use of the service, including the free tier, is subject to those terms, not this license. This repository does not contain the server's source code.

Links

• Website — opencapital.sh
• Connect guides — opencapital.sh/connect-ai
• Model Context Protocol — modelcontextprotocol.io