Case study
MyntAI: An Agentic Market Intelligence Layer Built on Claude
Mynt is a fintech and retail trading platform in the United Arab Emirates. ReapMind built MyntAI, its in-app market intelligence layer, over 14 weeks with a team of six.
12 min → 40s
Time to a researched view
+34%
Daily active users
Engagement at a Glance
Mynt
Client
Fintech · UAE
Sector
14 weeks
Engagement
Team of six
Delivery team
Claude + MCP
Core technology
MyntAI
Product
The Challenge
Mynt's users needed more than charts. A market decision depends on price action, macroeconomic conditions, breaking news, volatility, liquidity and sentiment at the same time. Those inputs lived in separate places, and a standard AI chatbot could not reach any of them.
In practice a Mynt user opened the app to check a position, then left it. They moved between the charting view, two or three news sites and a separate calculator, and reassembled the picture themselves. Mynt measured this at roughly twelve minutes for a single instrument, and users frequently abandoned the process partway through.
A conventional model could not close that gap either. It answers from a fixed knowledge snapshot, which goes stale the moment a central bank speaks or a company files an announcement. For a trading platform, that is the whole problem.
The Solution
ReapMind built MyntAI, an agentic intelligence layer in which Claude reasons across live market information rather than a static dataset.
Deterministic computation stays inside dedicated services. Claude interprets those outputs and relates them to the broader question being asked. That separation gives the system the flexibility of a language model with the repeatability of purpose built analytics.
Claude's reasoning
Decomposes a request into analytical tasks, selects the right tools, evaluates the retrieved information and synthesizes multiple signals into a coherent market view.
Web search
Pulls current information the model was never trained on, including central bank communication, corporate announcements, economic releases, and regulatory or geopolitical developments.
MCP and custom tools
Provide a standardized interface exposing Mynt's proprietary data pipelines, technical analysis functions, volatility calculations, correlation analysis, instrument metadata and historical datasets.
How a Request Flows
A single analysis moves through a defined pipeline rather than a single prompt and response:
Because retrieval happens at analysis time, the reasoning process can incorporate information that emerged after the model's training data. The system also separates observable market data from model generated interpretation, so a user can see which part of an answer is a measured signal and which part is context.
- Request
- Intent analysis
- Tool selection
- Data retrieval
- Market computation
- Contextual enrichment
- Claude reasoning
- Signal interpretation
- Response generation
The Outcome
Measured across the first three months after launch, against a baseline taken in the quarter before:
The modular design means new data providers and analytical services plug in through MCP without reworking the AI layer. Mynt has since added two further data sources without changes to the reasoning layer, each taking under a week from contract to production.
12 min → under 40s
Average time to a researched view on an instrument
+34%
Daily active users, three months after launch
~22,000
Queries handled through MyntAI each month
18
Proprietary Mynt tools exposed to Claude through MCP
6
External and internal data sources unified
-41%
Support tickets asking basic market questions
6.5s
Median response time for a multi-tool analysis
Why Claude
ReapMind evaluated three frontier models against the same set of MyntAI workloads before committing. Claude was selected on three grounds.
Reliability across multi-step tool chains
A single market question routinely triggers five to nine tool calls. In testing, Claude selected the correct sequence and recovered from failed calls more consistently than the alternatives, which mattered more than any single-response quality difference.
Native MCP support
MCP is Anthropic's own protocol, so exposing Mynt's eighteen internal tools required one standardized server rather than a bespoke integration per tool.
Restraint under uncertainty
In a regulated trading context, a model that separates measured signals from interpretation, and declines to assert what it cannot support, is worth more than a model that always produces a confident answer.
Compliance Note
MyntAI produces market analysis for information only and does not constitute financial advice. All analytical output presented to users is labelled as such, and the system distinguishes measured market data from model generated interpretation within every response.
Ready to build your next product?
Talk to our team about mobile apps, web platforms, and enterprise solutions tailored to your goals.