Build an AI Financial Agent with EODHD's MCP Server

EODHD has released an official MCP server with over 70 financial data tools, enabling AI agents to access fundamentals, prices, earnings, and news through a single interface. A developer demonstrates how to build a financial agent that answers multi-source questions in one call, eliminating the need for custom API wrappers.

Most people building financial AI agents connect four or five APIs by hand. One for prices. One for fundamentals. One for news. One for earnings dates. The problem isn't lack of data. It's fragmentation. If you're: this matters. Here's how to build an AI financial agent that pulls all four together, using EODHD's financial data API as the backbone. Here's what building a financial agent usually looks like with a plain REST API. You write a function to fetch prices. Another to fetch fundamentals. Another for news. Then you write orchestration logic so your agent knows which endpoint to call, in what order, and how to merge the responses into something the model can reason about. Each new data type means a new wrapper function. Each wrapper function is a new failure point. And every time the agent needs to answer a question that spans two data types, like "is this company's fundamentals strong enough to justify holding through earnings," you're the one stitching the context together, not the model. Developers usually discover this the hard way. The agent works fine in a demo with one data source. Then a real question comes in that needs prices, fundamentals, and news at once, and the whole orchestration layer has to be rebuilt. Financial agents don't need more APIs. They need a way to reach every data type through one consistent interface, so the model can decide what to call and combine it, instead of you hardcoding that logic in advance. That's what MCP Model Context Protocol is built for. MCP Model Context Protocol standardizes how an AI model discovers and calls external tools. Instead of you writing a custom wrapper for every endpoint, the model sees a list of available tools with their schemas, and decides which ones to call based on the question it's answering. It's the same underlying idea as Claude function calling, just extended to a whole catalog of tools instead of one function at a time. For financial data, this matters because a single question rarely maps to a single endpoint. EODHD ships an official MCP server with over 70 tools covering fundamentals, historical and real-time prices, earnings calendars, news, and sentiment. You point your agent at the server, and it gets access to all of it without you writing a single API wrapper. Skip the wrapper functions EODHD's MCP server exposes 70+ financial data tools out of the box, ready for Claude and other AI agents. → Get your EODHD API key Let's build an agent that can answer a question spanning multiple data types in a single call: fundamentals, upcoming earnings, and recent price action for a given ticker. pip install anthropic You'll need an EODHD API token and an Anthropic API key. EODHD's MCP server is available as a hosted endpoint, so there's no server to run yourself. EODHD's MCP server has two versions: v1 takes your API key directly in the URL, v2 uses OAuth 2.1 for clients that support it like Claude Desktop . For a server-side script like this one, v1 is simpler. python import anthropic client = anthropic.Anthropic api key="YOUR ANTHROPIC API KEY" EODHD API KEY = "YOUR EODHD API TOKEN" response = client.messages.create model="claude-sonnet-4-6", max tokens=1500, messages= { "role": "user", "content": "Give me a quick pre-earnings health check on AAPL: " "key fundamentals, the next earnings date, and how the " "stock has moved over the last month." } , mcp servers= { "type": "url", "url": f"https://mcp.eodhd.com/v1/mcp?apikey={EODHD API KEY}", "name": "eodhd-mcp" } for block in response.content: if block.type == "text": print block.text The mcp servers parameter is the whole trick. You're not calling three endpoints and merging the results yourself. You're handing the model a connection to EODHD's tool catalog and letting it decide what it needs. For this one prompt, the model will typically: Three tool calls, one conversation, zero orchestration code written by you. The API returns a mix of content blocks: text for the model's answer, mcp tool use for each tool it called, and mcp tool result for the raw data. If you want to inspect what the agent actually pulled useful for debugging or logging , filter by block type instead of assuming a fixed order: tool calls = {"name": b.name, "input": b.input} for b in response.content if b.type == "mcp tool use" for call in tool calls: print call "name" , call "input" This is worth doing early. It shows you exactly which tools the model reached for, which helps you spot when it's calling something unnecessary or missing a data type your prompt implied. The MCP agent above decides on its own which EODHD tools to call. That's convenient, but it also hides what's happening. If you want to actually understand and customize the logic behind "which stocks look like good opportunities," it helps to call the same EODHD endpoints directly and build the scoring yourself. This is also useful if you want a scheduled script that scans the market every morning instead of waiting for you to ask a chat agent. Let's build that: a short pipeline that pulls candidates from the Screener API, pulls fundamentals and recent price action for each one, checks the news sentiment, and combines all of it into a single opportunity score. Instead of pulling fundamentals for thousands of tickers, start narrow. The Screener API filters the entire market down to a shortlist in one request. python import requests API TOKEN = "YOUR EODHD API TOKEN" def get candidates : url = "https://eodhd.com/api/screener" params = { "api token": API TOKEN, "sort": "market capitalization.desc", "filters": ' "market capitalization"," ",1000000000 ,' ' "exchange","=","us" ,' ' "sector","=","Technology" ', "limit": 20, "fmt": "json", } response = requests.get url, params=params return response.json "data" candidates = get candidates tickers = c "code" for c in candidates print tickers This pulls US tech stocks above $1B market cap, sorted by size. Swap the filters for whatever criteria define a "good price" to you: low P/E, high dividend yield, positive EPS growth, or a 52-week low signal. Once you have a shortlist, fetch the fundamentals for each ticker. This is where you check whether the price is actually backed by solid financials, not just cheap for a reason. python def get fundamentals symbol : url = f"https://eodhd.com/api/fundamentals/{symbol}.US" params = {"api token": API TOKEN, "fmt": "json"} data = requests.get url, params=params .json highlights = data.get "Highlights", {} return { "pe ratio": highlights.get "PERatio" , "peg ratio": highlights.get "PEGRatio" , "profit margin": highlights.get "ProfitMargin" , "eps growth": highlights.get "EPSEstimateNextYear" , } Fundamentals tell you if a company is healthy. Price history tells you if the market has already priced that in, or if there's a gap worth paying attention to. python import pandas as pd from datetime import date, timedelta def get price trend symbol : url = f"https://eodhd.com/api/eod/{symbol}.US" params = { "api token": API TOKEN, "period": "d", "order": "d", "from": date.today - timedelta days=90 .isoformat , "fmt": "json", } prices = requests.get url, params=params .json df = pd.DataFrame prices last close = df.iloc 0 "close" month ago close = df.iloc 21 "close" if len df 21 else df.iloc -1 "close" change pct = last close - month ago close / month ago close 100 return {"last close": last close, "change 30d pct": round change pct, 2 } A stock with strong fundamentals that just dropped 15% in a month is a very different opportunity than one that already ran up 40%. Price and fundamentals don't tell you why a stock moved. Sentiment fills that gap, and it's often what separates a real opportunity from a value trap. python def get sentiment symbol : url = "https://eodhd.com/api/sentiments" params = { "s": f"{symbol}.US", "api token": API TOKEN, "from": date.today - timedelta days=14 .isoformat , "fmt": "json", } data = requests.get url, params=params .json entries = data.get f"{symbol}.US", if not entries: return {"avg sentiment": None} avg = sum e "normalized" for e in entries / len entries return {"avg sentiment": round avg, 3 } This is the part a raw API can't do for you. Each endpoint gives you one dimension. Deciding what a "good opportunity" means is a judgment call, and it belongs in your code, not buried in someone else's black-box score. python def score opportunity symbol : fundamentals = get fundamentals symbol price = get price trend symbol sentiment = get sentiment symbol score = 0 if fundamentals "pe ratio" and fundamentals "pe ratio" < 25: score += 1 if fundamentals "profit margin" and fundamentals "profit margin" 0.10: score += 1 if price "change 30d pct" < -5: score += 1 if sentiment "avg sentiment" and sentiment "avg sentiment" 0.15: score += 1 return { "symbol": symbol, "score": score, fundamentals, price, sentiment, } results = score opportunity t for t in tickers ranked = sorted results, key=lambda r: r "score" , reverse=True for r in ranked :5 : print r "symbol" , "score:", r "score" , "| P/E:", r "pe ratio" , "| 30d change:", r "change 30d pct" , "%", "| sentiment:", r "avg sentiment" Four data points, one loop, one ranked list. Reasonable valuation, healthy margins, a recent dip, and improving sentiment together are a much stronger signal than any one of them alone. This scoring logic is intentionally simple. You'll want to weight it differently depending on your strategy: a value investor cares more about P/E and margins, a swing trader cares more about the price drop and sentiment shift. The point is that the four EODHD endpoints give you the raw material, and the decision logic is yours to tune. None of this is financial advice. It's a framework for turning scattered data into a shortlist worth researching further, not a signal to buy. Once this scoring pipeline works as a standalone script, you can expose it to your MCP agent as a custom tool alongside EODHD's built-in ones. Then a prompt like "find me tech stocks that dropped recently but still look fundamentally solid" runs your exact scoring logic instead of the model guessing at criteria on its own. That's the real advantage of combining direct API calls with MCP: you get full control over the decision logic, and the model still handles the natural-language layer on top of it. This is, in practice, what it means to build an AI financial agent: not a single clever prompt, but a data layer you trust plus a model that knows when to reach for it. Once the agent can combine fundamentals, prices, earnings, and news on its own, the use cases stop being single-question demos. None of this requires new wrapper code. It requires better prompts and, occasionally, a narrower system prompt telling the agent which tools to prioritize. ❓ Do I need to run my own MCP server to use EODHD's tools? ✅ No. EODHD hosts the MCP server, so you connect to it with a URL and your API token, the same way you'd call a REST endpoint. ❓ Does MCP replace the EODHD REST API? ✅ No, it sits on top of it. The REST API still powers every tool call, MCP just gives the AI model a structured way to discover and use those endpoints without custom wrapper code. ❓ Can I limit which tools the agent has access to? ✅ Yes. You can scope the system prompt or restrict the conversation to specific tool categories if you don't want the agent reaching for endpoints outside a given use case, like keeping it to fundamentals and earnings only. ❓ Is this approach only useful with Claude? ✅ No. MCP is an open protocol, so EODHD's server works with any MCP-compatible client, including ChatGPT, Cursor, and Windsurf, not just Claude. ❓ How many API calls does a scan like this use? ✅ It depends on your shortlist size. Each screener request counts as 5 calls, and each fundamentals, price, or sentiment call for a ticker counts as 1. Scanning 20 candidates works out to roughly 65 calls, well within EODHD's free tier for occasional runs. ❓ Can I use this scoring script to automate actual trades? ✅ The script only ranks and prints candidates. Connecting it to a broker's execution API is a separate step, and it's worth adding manual review before any live-money decision, no matter how the shortlist was generated. If you're a software or API company looking to explain your product through high-quality educational content not marketing fluff , feel free to connect with me on LinkedIn. Want more content like this? Python, APIs, fintech, and AI agent tutorials for developers. → Visit kevinmeneses.com Looking for technical content for your company? I can help — LinkedIn · kevinmenesesgonzalez@gmail.com