The Equibles MCP server lets an AI agent pull live financial data straight into the conversation. Point any MCP-capable client at the server URL and your assistant can search filings, read holdings, and analyse earnings calls with no glue code. Every tool is available on every plan; the live-quote and option-chain tools need a paid plan for their market data and answer a Free key with a note naming the plan.
Use it as a financial data MCP server for Claude, ChatGPT, Grok, Cursor, VS Code, or any MCP client. It covers SEC filings, stock fundamentals, 13F holdings, earnings calls, options, congressional trades, short data, and FRED economic series through one remote endpoint.
https://mcp.equibles.com/mcp
Connected but unable to retrieve data? See connection troubleshooting for the optional
?auth=oauthsetup URL. Follow the Grok guide for Grok; keep the default URL for ChatGPT, Claude, and other working clients.
Choose your client
Each client connects a little differently. Pick yours for setup, verification, and troubleshooting tailored to it.
mcp.json config.Set up Cursor →
VS CodeUse the built-in Copilot MCP configuration.Set up VS Code →
WindsurfAdd Equibles to Cascade's MCP config.Set up Windsurf →
Google AntigravityNo-code setup for your Google agents.Set up Antigravity →
Gemini CLIConnect through your terminal settings.Set up Gemini CLI →
Any MCP clientUse the generic Streamable HTTP recipe.View generic setup →
Authentication
Most modern clients can sign in over OAuth — no key to copy or store. Any client can also use an API key, sent as a Bearer header or an ?api_key= query parameter on the URL. See Authentication for details.
Compact output (GCF)
Many table answers come back as markdown tables by default. Ask for Graph Compact Format instead and the column names are factored into one header, with each row a single pipe-delimited line.
If your client can set headers, send X-Equibles-Output-Format: gcf. If it signs in with an API key in the URL, add the format beside the key instead:
https://mcp.equibles.com/mcp?api_key=eq_...&output_format=gcf
Prefer the header when you sign in over OAuth, and keep the server URL from your client's setup guide, including ?auth=oauth if that guide requires it.
The cell values are identical either way, the same figures and the same formatting, and any answer falls back to a markdown table whenever GCF would not be shorter, so nothing is ever lost to it. Measured on real answers from the server, the answer text runs 7 to 13% shorter: the table compresses and the prose around it does not, so a wide table of many short values saves most and a small table inside a long preamble saves least. That is the text your client hands the model rather than the bytes on the wire, which can grow, because JSON escapes every quotation mark the compact form adds. The setting applies to that one request, so turning it on for one client changes nothing for your others.
Which answers use it. Today the compact form covers short volume and short interest, off-exchange volume, fails-to-deliver, congressional trades and member disclosures, FRED indicators and the economic calendar, CFTC positioning, CBOE put/call and VIX history, FDA advisory meetings, government contracts, and institution and adviser search. Prices, financial statements, insider trades, dividends, 13F holders and portfolios, fund filings and filing search still answer in markdown whatever you ask for, because those answers continue past the table with footnotes and truncation notes. Extending the compact form to them is on the list.
Ask for Equibles by name
Connecting the server is only half of it. Most assistants decide for themselves where an answer comes from, and given a market question they will often answer from training data or run a web search instead of calling a tool. So make the ask explicit:
Two clients also need the server switched on for the conversation, not just connected in settings:
- Claude: enable the Equibles connector for the chat, and name it in the prompt so Claude uses the tools rather than its own web search. See Claude web, Claude Code or Claude Desktop.
- ChatGPT: pick Equibles from the composer's + menu in each new chat, or name it in the prompt. See ChatGPT.
Naming Equibles is what gets you the processed data: every figure carries the filing, the period, and the form it came from, which a web search result does not.
What your agent can do
Once connected, just ask in plain English — the model picks the right tool. A few to try, spanning what the server covers:
For focused workflows, start with the price and technical tools, fundamentals and valuation tools, SEC filing search tools, insider trading tools, short-data tools, or options data tools. Each family page lists the available tools, parameters, example prompts, and sample output; the tools overview maps every other task.
Fundamentals & filings
Ownership & flows
Markets & screening
Browse every tool — each with its parameters, an example prompt, and a real sample of its output — in the tool reference. To string several together, follow a recipe; to bake in our research conventions, add the Claude Skill. Every call counts against your daily limit.
Troubleshooting
Your agent answered without calling Equibles?
If the numbers look plausible but the assistant never shows a tool call, it's answering from its own training data or from a web search instead of live data, and this is the single most common issue. Start the prompt with "using Equibles, ...". If it still won't, confirm the server is enabled for that conversation, not just connected in settings, on your client's page.
Access denied or over your limit?
Re-check your API key or OAuth sign-in on your client's setup page. If you've used up today's calls, the server replies with a message explaining the limit — see Rate limits to raise it.