Skip to content
Epiphany Dynamics Epiphany Dynamics

How a Research Team Got Claude Answering Live Analyst-Ratings Questions in About 36 Hours, Three Days Ahead of Deadline, With Nothing Bolted On

The team wanted to ask Claude plain-English questions about upgrades, downgrades, price targets, and consensus, and get answers pulled live from Benzinga. No front end. One data source. A hard deadline. Epiphany Dynamics built exactly that, verified it against the live API the next morning, and held scope where the client set it.

A boutique research team case study: LIVE RATINGS. ANSWERS IN THIRTY-SIX HOURS.

Analyst questions answered. No front end.

Client
BR

Boutique Research Team

Built
MCP server exposing analyst-ratings tools to Claude Desktop
Sector
AI Agent Deployment

A boutique research team. No company name, by request.

Three questions an analyst asks
  1. 01 What changed on this ticker today? The answer should come back as a short narrative of rating moves, not raw JSON. Illustrative question, included to show the shape of the tool, not an actual client query.
  2. 02 How has consensus moved over the last week, and where do individual analysts sit against it? Illustrative question.
  3. 03 Who upgraded or downgraded this name recently, and what price target did each one set? Illustrative question.
Five tools

  • 01 Analyst ratings

    The current rating and rating history for a ticker, shaped into structured data Claude can read.

  • 02 Rating changes

    The upgrade and downgrade feed, so a question about what moved comes back as a narrative.

  • 03 Price-target changes

    New targets as they are set, with the analyst behind each one.

  • 04 Consensus

    The aggregate view across the analysts covering the name.

  • 05 Recent activity

    What changed on a ticker today, assembled from the tools above.

Contract to verified build ~36 hours Offer accepted Friday at 9:30 pm; all five tools verified against the live API the next morning
Against the deadline 3 days early Delivered Saturday against the client's hard deadline the following Tuesday
Lines of front-end code 0 No front end, no dashboard; the two additional sources the client declined were not added
Field Note

The things we did not build

Two more data sources would have been easy to add. They were declined, so they were not added. A research tool the team can reason about beats a bigger one nobody asked for.

FAQ
Why is there no front end?
The client explicitly set that scope: no front end, no dashboard, no extra data sources. The team already works inside the Claude Desktop app, so the MCP server runs there, with install paths also documented for Claude Code and a direct SDK.
What can Claude actually ask?
The server exposes five discrete tools covering analyst ratings, price-target changes, consensus, and recent upgrades and downgrades against the live Benzinga API. Plain-English questions like "what changed on this ticker today" route to the right tool and come back as a narrative, not raw JSON.
How does this compare to a data-terminal seat?
The Benzinga data feed the tool queries costs roughly $1,200 a year versus roughly $32,000 a year for a single terminal seat, about 27x less (sources: massive.com/partners/benzinga, costbench.com terminal pricing). The comparison is deliberately narrow: a terminal does far more, but for this one job, conversational analyst-ratings lookups, the team did not need one.

Want Your Team Asking Questions of Your Own Data Instead of Pasting It Into a Chat Window?

Start with one source.

Book my free audit call The service → All case studies →

Free 30-minute audit. No obligation, no pitch. Or take the 2-minute AI quiz.