接入你的智能体
從發布者安裝。不是 PlugMCP 端點。
遠端 URL 還是本機指令? 兩種安裝方式的差異
來自發布者
Self-hosted, open-source financial data MCP server for AI agents — SEC filings, XBRL financials, 13F holdings, insider & congressional trades, short interest, FRED, CFTC/CBOE and daily prices across 64 MCP tools. Equibles Cloud adds earnings call transcripts, live quotes, options chains with Greeks, guidance and a screener.
Equibles is a self-hosted, open-source financial data MCP server — an open-source alternative to a Bloomberg Terminal, built for AI agents rather than humans. It scrapes, stores, and serves SEC filings and XBRL financials, 13F institutional holdings, insider and congressional trades, FINRA/SEC short data, FRED economic indicators, CFTC and CBOE positioning, fund filings, government contracts, and daily stock prices, and exposes them over the Model Context Protocol, so Claude, ChatGPT, Cursor, or any agent can query them directly. Its stock universe covers US listings plus verified listings on the major European exchanges. Runs on your own hardware with Docker, for free, forever.
This is the open-source core of Equibles. Equibles Cloud runs this exact core and adds more tools on top: earnings call transcripts and audio, real-time quotes, options chains with Greeks, LLM-extracted KPIs and guidance, buybacks, IPO filings, executive changes, valuation multiples, index composition, portfolio tracking, and a US-market screener. It also lists Chinese and Hong Kong shares beside the European ones. Same protocol, same tool names, nothing to run — see what's included.
Don't want to run anything? Point your AI assistant at
https://mcp.equibles.com/mcpand get a free API key at equibles.com/mcp — 100 requests/day, no card. Per-client setup guides live at daniel3303/stock-market-mcp-server.
See docs/ for the user guide and technical documentation.
Everything marked Self-hosted is scraped, stored, and served by this repo — 68 MCP tools, no account, no key. Equibles Cloud runs this exact core over the same protocol with the same tool names, and adds more tools on top.
Nothing above is held back from this repo. The cloud-only rows are data whose production needs infrastructure you would not stand up at home:
Beyond the data, the cloud is already backfilled and kept current, and it is managed — no scrapers to babysit. This repo is free forever under AGPL-3.0; the cloud has a free tier at 100 requests/day with no card.
Start on the free tier, and self-host whenever you'd rather own the stack.
The fastest way to get everything running. Requires Docker.
git clone https://github.com/daniel3303/Equibles.git
cd Equibles
cp .env.example .env
# Edit .env and set SEC_CONTACT_EMAIL (required by SEC EDGAR fair access policy)
docker compose up
This starts:
Data scraping starts automatically. SEC filings, holdings, insider trades, and congressional trades will begin populating within minutes.
All settings can be configured via a .env file in the project root (recommended for Docker) or environment variables.
FINRA Short Data (free API key required):
The FINRA scraper (short volume and short interest) requires a free API key. Without it, the scraper skips gracefully and all other scrapers run normally. Fails-to-deliver data comes from SEC and works without FINRA credentials.
To get a key:
Finra__ClientId and Finra__ClientSecret in your .env file or environment variablesThe older
developer.finra.org"Teams & Apps" flow has been retired — use the API Console above.
FRED Economic Data (free API key required):
The FRED scraper requires a free API key from the Federal Reserve Bank of St. Louis. Without it, the scraper skips gracefully and all other scrapers run normally.
To get a key:
Fred__ApiKey in your .env file or environment variablesTicker Filtering (optional):
By default, all tickers are synced. To limit data syncing to specific stocks, set a single ticker list that applies to all scrapers:
# .env — sync only these tickers (applies to all scrapers)
Worker__TickersToSync__0=AAPL
Worker__TickersToSync__1=MSFT
Worker__TickersToSync__2=GOOGL
When not set, all stocks are synced.
Minimum Sync Date (optional):
By default, all scrapers start from January 2020. Set a more recent date for faster initial sync, or go as far back as 2000-01-01 for more historical data:
# .env — start syncing from 2024 instead of 2020
Worker__MinSyncDate=2024-01-01
Embedding (opt-in):
Update notifications (optional):
Authentication (optional):
When set, the web portal requires login and the MCP server requires Authorization: Bearer <key> header. When unset, everything is open access (default).
The web portal checks GitHub Releases on a schedule and shows a banner when a newer version is available (disable with CHECK_FOR_UPDATES=false). To update to the latest release:
Docker Compose:
git pull
docker compose up -d --build --remove-orphans
Include any optional Compose override files in the upgrade command so those services remain enabled.
From source:
git pull
dotnet build Equibles.sln
Database migrations are applied automatically on startup. Review the changelog for notable changes before upgrading.
The web portal at http://localhost:8080 provides a browser-based interface for exploring data:
/advisers), ranked by assets under management, with per-firm profiles (regulatory AUM, main office, employee count, fee structure)/most-shorted, ranked by FINRA short interest) and largest daily short volume (/short-volume), each with a date selector, server-side sort, and paginationThe MCP server exposes financial data tools for AI assistants (Claude, ChatGPT, etc.):
By default the table tools return markdown tables. Graph Compact Format
is used instead when it is smaller, otherwise it falls back to markdown. GCF carries the same
cells the markdown table would show, with the column names factored into a single header and
the per-cell | padding and separator row dropped.
It reaches the answers rendered through MarkdownTable.Render, which today is the short-volume
and short-interest tools, off-exchange volume, fails-to-deliver, congressional trades and member
disclosures, the FRED series and calendar tools, CFTC positioning, CBOE ratios, FDA advisory
meetings, government contracts, and the institution and adviser search tools. Tools that assemble
their answer with MarkdownTable.Start and append rows themselves are not encoded, because
those answers continue past the table with truncation notes and footnotes; that includes the
price, financial-statement, insider, dividend, 13F-portfolio, fund-filing and filing-search
tools. Routing them through the shared renderer is tracked separately.
Ask for it per request, which is what a shared server wants, with either the
X-Equibles-Output-Format: gcf header or an ?output_format=gcf query parameter (the query
parameter is there for clients whose only configuration is a URL). markdown is accepted the
same way, so a caller can opt out of a server that defaults to GCF. An unrecognized value is
ignored rather than refused, and the server's own default applies.
Set EQUIBLES_OUTPUT_FORMAT=gcf to make GCF the process-wide default instead. A request
that names a format wins over it; one that names nothing gets it.
Per-request selection relies on each tool call arriving as its own HTTP request, which is how the streamable HTTP transport works in both its stateless and session modes: the format is read off that request and applies to it alone. A transport that instead dispatched calls on a connection opened earlier would carry whatever the opening request asked for.
How much it saves depends on the shape of the answer. The table compresses, the prose around it does not, so a small table inside a long preamble saves least and a wide table of many short values saves most. Measured end to end against a deployed server on real answers from three of the covered tools, 30 sessions of daily short volume, 25 congressional trades and 24 monthly observations of a FRED series, the answer text is 7 to 13% shorter in characters.
That is the text your client hands the model, not the HTTP payload. JSON escapes every quotation mark the encoder adds, so the transport bytes can grow on the same answer whose text shrank.
It is deliberately conservative: every cell value is preserved verbatim (the compact-USD, comma-grouped, adaptive-decimal and em-dash formatting is untouched — GCF only changes the framing, not the numbers a model reads), and any row that does not match the header's column shape falls back to the markdown table, so a result is never dropped or garbled. Output is unchanged for a caller that asks for nothing on a server that sets nothing. GCF is MIT-licensed and its encoder is zero-dependency.
Add this to your Claude Desktop config file (claude_desktop_config.json):
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"equibles": {
"url": "http://localhost:8081/mcp"
}
}
}
Restart Claude Desktop and the Equibles tools will be available. You can then ask questions like "Who are the top institutional holders of AAPL?" or "Search Apple's latest 10-K for revenue growth discussion."
Add the MCP server to Claude Code:
claude mcp add equibles --transport http http://localhost:8081/mcp
Add this to your ChatGPT Desktop config file:
macOS: ~/Library/Application Support/com.openai.chat/mcp.json
Windows: %APPDATA%\com.openai.chat\mcp.json
{
"servers": {
"equibles": {
"url": "http://localhost:8081/mcp"
}
}
}
Restart ChatGPT Desktop and the Equibles tools will be available.
In OpenClaw, add an MCP server with the URL http://localhost:8081/mcp (HTTP transport).
Any MCP-compatible client can connect to http://localhost:8081/mcp (HTTP transport).
Clients that launch a server as a subprocess can run it over stdin/stdout instead of HTTP:
dotnet build src/Equibles.Mcp.Server
dotnet run --no-build --project src/Equibles.Mcp.Server -- --stdio
Build first: a plain dotnet run restores and prints build output on stdout, which corrupts the protocol stream. The server reads ConnectionStrings:DefaultConnection as usual, so point it at the same database the worker fills. Logs go to stderr. The API key and the per-request output-format header and query parameter do not apply; set EQUIBLES_OUTPUT_FORMAT=gcf in the subprocess environment instead.
This self-hosted build exposes 68 tools over MCP. The hosted server at https://mcp.equibles.com/mcp runs this same core and adds more on top. Full catalog and client setup: daniel3303/stock-market-mcp-server.
13F institutional holdings
Insider trading
SEC filings search
Funds, ETFs & advisers
Fundamentals (XBRL)
Economic data (FRED)
Futures (CFTC COT)
Volatility (CBOE)
FDA calendar
Congressional trading
Short data (FINRA)
Stock prices
Government contracts (USAspending)
Vector embeddings enable semantic search over SEC filings (e.g., "find revenue growth discussion in Apple's 10-K"). This requires downloading the Ollama runtime (~2GB) and the Qwen3-Embedding-0.6B model (~640MB).
docker compose -f docker-compose.yml -f docker-compose.embedding.yml up
This adds Ollama and configures the existing web, MCP, and worker services to use it:
Without the embedding override, BM25 full-text search via ParadeDB still works out of the box — vector search is purely additive.
Bundled Ollama is the default because it needs no GPU and runs anywhere. For bulk embedding at scale, point Embedding__Provider=OpenAI at a batched server such as vLLM or Text-Embeddings-Inference — they continuously batch on the GPU and are far faster than Ollama for large corpora. See docs/guide/how-to-use-external-embedding-endpoint.md.
See CONTRIBUTING.md for development setup, project architecture, and how to extend the platform. Contributors must sign the Contributor License Agreement — this is handled automatically by a bot when you open a pull request.
Daniel Oliveira
Self-hosted, open-source financial data MCP server for AI agents — SEC filings, XBRL financials, 13F holdings, insider & congressional trades, short interest, FRED, CFTC/CBOE and daily prices across 6
總覽
Self-hosted, open-source financial data MCP server for AI agents — SEC filings, XBRL financials, 13F holdings, insider & congressional trades, short interest, FRED, CFTC/CBOE and daily prices across 64 MCP tools. Equibles Cloud adds earnings call transcripts, live quotes, options chains with Greeks, guidance and a screener.
工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
這筆記錄未列出必要的環境變數。
Equibles 透過 stdio 在本機執行。
上次檢查時間:2026-11-17。 Self-hosted, open-source financial data MCP server for AI agents — SEC filings, XBRL financials, 13F holdings, insider & congressional trades, short interest, FRED, CFTC/CBOE and daily prices across 64 MCP tools. Equibles Cloud adds earnings call transcripts, live quotes, options chains with Greeks, guidance and a screener.