Build an AlphaSense-Style Research Assistant with MCP

Connect Claude Code, Cursor or Claude Desktop to filings, statements, quotes and macro data over MCP, see what a research question costs in data calls, and see the parts of AlphaSense no public API can replace.
TL;DR
- •With one MCP server, Claude can pull SEC filing text, US financial statements, quotes, FRED and BLS series, and prediction-market odds. That covers the public-data half of an AlphaSense or Fiscal.ai workflow.
- •It cannot replace broker research, expert-call transcripts or private-company data. Those are licensed content, and AlphaSense, PitchBook and FactSet sell them only with a seat.
- •A typical five-call research question costs about 610 apipick credits ($0.61) in data, plus model tokens. At 100 questions a month that is about $61, more than a $39 to $49 subscription to an AI research terminal.
- •Build when the assistant must live inside your own product or workflow. Buy when you just want answers.
Can you build an AlphaSense-style research assistant yourself?
Yes, for the public-data half. Claude connected to a finance MCP server can read SEC filings, US financial statements, quotes, macro series and prediction-market odds, then answer a research question with sources. No public API replaces the other half, AlphaSense's licensed broker research and expert-call transcripts. This guide shows the setup, what one question costs, and where buying is the better decision.
We build apipick, one of the MCP servers used here. The cost figures use our list prices and calls we ran on 2026-11-16. The competitor facts come from each vendor's own pages, linked at the end.
What does AlphaSense have that an API cannot give you?
Licensed content. AlphaSense says it searches more than 500 million documents, including broker research from over 1,000 firms and expert-call transcripts (it completed its acquisition of Tegus, an expert-call library). None of that is available through a public data API at any price.
| AlphaSense | Fiscal.ai terminal | DIY: Claude + MCP | |
|---|---|---|---|
| SEC filings and statements | Yes | Yes | Yes |
| Earnings-call transcripts | Yes | Yes | Only with a transcript API (FMP Ultimate, Quartr, Alpha Vantage) |
| Broker (sell-side) research | Yes, 1,000+ firms | No | No |
| Expert-call transcripts | Yes (Tegus) | No | No |
| Private companies | Private-market intelligence product | No | Only with a PitchBook or CB Insights licence |
| Macro series and prediction markets | Not checked | Not checked | Yes: FRED, BLS, World Bank, IMF, Polymarket, Kalshi |
| Runs inside your own app | Enterprise Suite API and MCP | API and MCP, sales-priced | Yes |
| Price | Contact sales | Pro listed at $49/month in its help center | Pay per call plus model tokens |
If your research depends on sell-side notes or channel checks from expert calls, stop here: buy AlphaSense, or pair a terminal with a broker-research source. The rest of this guide is for workflows built on public filings and market data.
How do you connect Claude to finance data over MCP?
Add one remote MCP server to Claude Code, Cursor, VS Code or Claude Desktop, and the model sees each data source as a tool it can call. The steps below use apipick's server at https://www.apipick.com/mcp. The same pattern works with Fiscal.ai, Financial Datasets or FMP, which document their own MCP endpoints.
1Get an API key
Create a key in the dashboard. New accounts get 100 free credits, enough for a few research questions. Only successful calls are charged.
2Add the MCP server to your client
In Claude Code, run:
claude mcp add --transport http apipick https://www.apipick.com/mcp \
--header "Authorization: Bearer $APIPICK_API_KEY"For Cursor, add this to ~/.cursor/mcp.json:
{
"mcpServers": {
"apipick": {
"url": "https://www.apipick.com/mcp",
"headers": { "Authorization": "Bearer YOUR_API_KEY" }
}
}
}The MCP server page has the VS Code and Claude Desktop versions.
3Check which finance tools the model sees
The finance tools are markets_search, financials_search, sec_search, economic_search, earnings_calendar_search and prediction_markets_search. Each tool description includes its credit cost, so the model can see what a call costs before it makes one.
4Give the assistant a research brief
A short system prompt keeps the model on sources and away from mental arithmetic:
You are an equity research assistant.
- Read the primary filing (sec_search) before quoting any figure from the latest quarter.
- Cite the accession number or source for every number.
- Do arithmetic in code or show the inputs; never estimate a ratio from memory.
- If a tool returns nothing, say so instead of guessing.Which tool calls does one research question use?
A question such as "How did Microsoft's last quarter compare with the one before, and what does the macro backdrop look like?" maps to about five calls. Below is the plan we would expect the model to follow, with the cost and the latency we measured for each call on 2026-11-16.
| Tool | Credits | Measured latency | |
|---|---|---|---|
| Current price | markets_search | 120 | 1.8 s |
| Last six quarters of the income statement | financials_search | 200 | 1.1 s |
| Latest 10-Q text | sec_search | 120 | 5.6 s |
| Previous 10-Q or 8-K | sec_search | 120 | 2.5 to 7.7 s |
| CPI series | economic_search | 50 | 3.1 s |
| Total | 5 calls | 610 ($0.61) | about 15 to 20 s sequential |
One thing the plan has to handle: in our test, the statements response stopped at the quarter ending 2026-06-30, while Microsoft's September-quarter 10-Q had been filed on 2026-10-28 and sec_search returned it. That is why the system prompt tells the model to read the filing for the latest quarter. Filing search also needs the company and form in the query ("Microsoft latest 10-Q"). A topic query such as "Tesla 10-K risk factors" returned nothing, so have the model fetch the filing and search the text itself.
Is building cheaper than buying a research terminal?
For one analyst asking questions, no. At about $0.61 in data per question, 100 questions a month costs about $61 before model tokens, against $39 to $49 a month for AI research terminals such as PickSkill Pro or Fiscal.ai Pro.
Building pays off in other situations:
- The assistant lives inside your product. A terminal subscription cannot be embedded in your app; an API and MCP server can.
- You combine market data with your own documents, such as internal memos, CRM notes or a data room, which no terminal can see.
- You need a fixed, auditable workflow, for example a screening step that runs every morning and writes to your database.
- Usage is occasional. Ten questions a month cost about $6 in data, with no subscription.
If you mostly want answers, decks and portfolio monitoring, buy a tool. Our sister product PickSkill is an AI analyst for US, Hong Kong and A-share stocks that exports sourced PowerPoint, Word and Excel files, from $15 a month. For a broader list of terminals by budget, see the AlphaSense alternatives comparison on its blog.
Which finance MCP servers work without an enterprise contract?
In November 2026, the incumbents (Bloomberg, FactSet, S&P Capital IQ, PitchBook, AlphaSense) all offer MCP access, but only on top of an existing enterprise licence. Self-serve MCP servers come from smaller vendors.
- Fiscal.ai: API and MCP at
api.fiscal.ai/mcp; free tier limited to 100 companies, paid limits via sales. - ROIC.ai: API and MCP on every plan, from $24 a month.
- StockInsights.ai: API with a 7-day trial, plus an MCP server, from $17 a month.
- Financial Datasets, FMP, sec-api.io, EODHD: MCP servers alongside their data plans. We compare their prices in Financial data APIs for AI agents.
- Quartr: MCP included with Quartr Pro, for transcripts, audio and slides. See Earnings call transcript APIs.
- apipick: one server for finance plus web, news, patents and research search, pay per successful call.
For a code-first version of this assistant with Python instead of MCP, see how to build an investment research agent, and for a filings-only workflow, building a due-diligence agent on SEC filings.
Sources, read on 2026-11-16: AlphaSense pricing, AlphaSense completes acquisition of Tegus, Fiscal.ai help center, Fiscal.ai API docs, Fiscal.ai MCP, ROIC.ai pricing, StockInsights.ai pricing, Quartr MCP, Bloomberg Enterprise MCP, FactSet MCP announcement, PitchBook connector for Claude, PickSkill pricing, apipick MCP server.
Frequently Asked Questions
Can I build my own AlphaSense with Claude?
You can rebuild the public-data half: SEC filings, financial statements, quotes, macro series and news, connected to Claude through an MCP server. You cannot rebuild AlphaSense's licensed content, such as broker research from more than 1,000 firms and expert-call transcripts, because those are only sold with an AlphaSense seat.
Does AlphaSense have an API or MCP server?
Yes, but only on its top tier. AlphaSense's pricing page lists an API, an Agent API and a SuperAnalyst MCP server as part of the Enterprise Suite, and prices are available from its sales team only.
How much does a DIY research assistant cost to run?
On apipick, a question that reads one quote, one statement set, two SEC filings and one macro series costs 610 credits, or $0.61, plus the language model's token cost. At 100 questions a month the data alone is about $61.
Which finance MCP servers can I use without an enterprise contract?
Self-serve options on 2026-11-16 include Fiscal.ai, ROIC.ai, StockInsights.ai, Financial Datasets, FMP, sec-api.io, EODHD and apipick. Bloomberg, FactSet, S&P Capital IQ, PitchBook and AlphaSense also offer MCP access, but only to existing enterprise customers.
Sarah Choy is the CEO of apipick. She writes about building production-ready APIs for AI agents and LLM workflows.