Building AI Trading Agents with MCP and altFINS

CryptoOne — Tue, 05 May 2026 13:49:28 +0000

A practical way to turn an LLM into a crypto research assistant is to connect it to a tool layer that can return real market data instead of guesses. The altFINS MCP server does exactly that by exposing crypto screeners, technical analysis, signals, news, calendar data, and portfolio data through the Model Context Protocol, which works with clients like Claude Desktop, VS Code, and Copilot Studio.

The stack becomes even more useful when MCP is paired with the altFINS CLI and API. altFINS positions the API as a developer-first crypto analytics platform with 150+ indicators, 130+ signals, and historical coverage across 2,000+ assets, while the CLI gives direct terminal access without building a client from scratch.

Why this setup works

A common problem with AI trading assistants is that they sound smart while operating on stale or invented data. The altFINS MCP server is designed as a structured, AI-friendly interface to live crypto analytics, which reduces hallucination risk by letting the model call actual tools for screener results, technical analysis, OHLCV data, news, and historical indicators.

That changes the workflow in a useful way. Instead of manually hopping between dashboards, copying values into prompts, or building an entire API wrapper before testing an idea, a developer can ask for a screen like “find coins with RSI below 30 and bullish MACD” and let the client invoke the relevant tool directly.

What was used

This workflow uses three parts that fit together well:

*altFINS MCP server * for natural-language access to crypto data and analytics through supported AI clients.
altFINS CLI for quick terminal checks, exports, and repeatable commands such as af markets search, af signals list, and af tui markets.
altFINS API for production use cases that need direct integration, rate limits, and predictable data access in custom apps or agent frameworks.

A useful pattern is to explore with MCP, validate with CLI, and then automate with the API. That keeps early experimentation fast while leaving a clean path toward a real bot, dashboard, or research workflow.

Connecting Claude to altFINS MCP

The MCP endpoint is available at https://mcp.altfins.com/mcp, and the docs show Claude Desktop as one of the supported entry points.

In that setup, the connection is commonly made through mcp-remote, with the API key passed in the X-API-Key header.

json
{
"mcpServers": {
"altfins-mcp-server": {
"command": "npx",
"args": [
"mcp-remote",
"https://mcp.altfins.com/mcp",
"--header",
"X-API-Key: ${API_KEY}"
],
"env": {
"API_KEY": "YOUR_ALTFINS_API_KEY"
}
}
}
}

After saving the config and restarting Claude Desktop, the docs recommend starting a new chat and testing it with prompts such as “Give me a complete technical analysis of SOL” or “Show me the latest crypto news about Ethereum.”

The tools that matter most

The strongest part of the integration is that the MCP server exposes a focused tool set rather than a generic text dump. According to the documentation, the server includes tools for Screener, Technical Analysis, OHLCV Data, Analytics History, Signal Feed, news, calendar events, and portfolio information.

That tool mix is a strong fit for agent workflows because it supports narrowing and expansion in the same conversation. A model can begin with market-wide discovery through the screener, drill into one asset with technical analysis, and then add context from recent news or upcoming events without switching systems.

Example 1: progressive market scan

One useful workflow is progressive narrowing. Start broad, then tighten the conditions as the model returns candidates.

"Find AI coins with bullish MACD and RSI below 50."

That kind of prompt maps naturally onto the screener because altFINS supports screening thousands of assets by indicators such as RSI, MACD, SMA, trends, patterns, categories, and exchanges.

A second prompt can then focus the search further.

"Take the top 5 and compare their trend strength, latest signals, and nearest resistance levels."

From there, the next step is not “which coin will moon,” but “which ones are worth more research today.” That framing keeps the agent useful as a research assistant instead of pretending it can guarantee outcomes.

Example 2: deep dive on a single asset

Once a shortlist exists, the technical analysis tool becomes the fastest way to create a compact analyst-style view. The altFINS docs explicitly show a Claude example for “Give me a complete technical analysis of SOL,” which is a good model prompt because it is specific enough to trigger the right tool and broad enough to return a usable decision snapshot.

"Give me a complete technical analysis of SOL."
Include trend, momentum, support and resistance, and any notable signals.
A follow-up prompt can make the result more operational:

Now turn that into a swing-trade research note with entry ideas, invalidation level, and what would confirm the setup.
The result is not a substitute for execution discipline, but it is a major speed upgrade for daily prep.

Example 3: validating ideas in the terminal

MCP is great for fast iteration, but the CLI is often the better tool when the goal is repeatability. The altFINS CLI is described as a terminal-native interface to the same analytics stack, with install steps such as brew install altfins-com/tap/af, followed by af auth set and af auth status.

bash
brew install altfins-com/tap/af

save API key and confirm access
af auth set
af auth status

quick market scan
af markets search --symbols BTC,ETH,SOL --display-type MARKET_CAP,RSI14,MACD

recent bullish signals
af signals list --direction BULLISH --from 2026-03-01

interactive terminal UI
af tui markets

This is especially useful when an MCP prompt surfaces an idea and there is a need to verify it outside the chat interface. The CLI also helps when exporting data for a notebook, script, or backtest workflow.

Example 4: using the API for production agents

The API is the next step when moving from exploration into an application or internal tool. altFINS describes the product as a production-grade crypto analytics API for trading bots, fintech apps, algo systems, and autonomous AI agents, with MCP serving as the model-friendly interface on top.

That separation is useful in practice. MCP is excellent for fast human-in-the-loop workflows, while the API is a better fit for scheduled jobs, dashboards, and backend services where direct control over requests and responses matters more than conversational interaction.

A simple real-world pattern looks like this:

Use MCP in Claude to discover a screening prompt that returns good candidates.

Recreate the same logic in CLI commands for quick verification and export.

Move the strategy into API-driven code for automation, alerts, or a dashboard service.

A practical daily workflow

A realistic way to use the stack is to treat it as a research pipeline rather than a black-box trading engine. The platform update notes from altFINS describe new 2026 support for signal bots, automated algos, dashboards, MCP integrations, and agentic workflows such as n8n, which fits this kind of modular workflow well.

A simple daily routine could look like this:

Morning: Use MCP to ask for breakouts, oversold rebounds, or category-specific setups.

Midday: Use the CLI to validate a shortlist and export data for additional checks.

Afternoon: Push the refined logic into API-driven automation for alerts or internal reporting.

This is not fully autonomous trading, and that is part of the appeal. It keeps judgment and risk management with the operator while removing a lot of repetitive data gathering.

Where this approach shines

This setup is strongest when speed of analysis matters more than custom infrastructure. altFINS says the API and MCP stack is built for trading apps, AI agents, bots, and fintech use cases, and the CLI extends that by giving terminal-native access to the same data layer.

Good fits include:

Market research assistants that answer with live indicator data instead of generic text.
Internal analyst copilots that turn watchlist questions into structured research notes.
Bot prototypes that need fast screening, signal lookups, and historical context before a heavier engineering build.
Content and reporting workflows that need fresh crypto analysis quickly, as shown in altFINS examples around Claude-based article generation.

Limits to keep in mind

The biggest risk is not technical. It is over-trusting a smooth interface. Even when the model is calling live tools, a smart workflow still needs verification, position sizing rules, and execution discipline, especially in crypto markets where conditions change fast.

There is also a practical distinction between exploration and production. MCP is excellent for conversational discovery, but long-term systems still benefit from direct API integrations, logging, monitoring, and explicit error handling.

Closing take

For developers experimenting with AI agents in crypto, altFINS is interesting because it offers three useful layers at once: a production-oriented API, a terminal-friendly CLI, and an MCP server that makes the analytics stack accessible inside modern AI clients.

The result is a cleaner path from idea to workflow. A prompt can become a screen, a screen can become a repeatable terminal command, and that command can become an automated system with much less glue code than a traditional build would require.

Forem: CryptoOne