Screen Earnings Calls Across the Market with MCP Tools

Most people read one earnings call at a time. They pull up a transcript, skim the prepared remarks, jump to the Q&A, and move on. That works when you already know the story you're checking. It falls apart the moment the real question is comparative: Who across my coverage is raising prices? Which semis names are guiding capex higher? Where are layoffs showing up first?

The power move in earnings research isn't reading a single call more carefully. It's screening earnings calls across companies at once, then narrowing to the handful that actually matter. That's a different workflow, and it needs a different tool. When you screen earnings calls with AI over a corpus that spans years, sectors, and geographies, you stop reacting to headlines and start spotting themes before the sell-side writes them up.

This guide shows how to do exactly that with MCP tools connected to EarningsCalls.dev — starting from a single prompt and scaling all the way to 500-ticker aggregate sweeps via the API.

Why full-text + speaker-tagged search matters

Cross-market screening only works if the underlying search is precise. Two properties make the difference.

Full-text search over the whole transcript. You're not matching on a summary or a headline — you're matching on what executives and analysts actually said, word for word. The corpus covers 2020 to present (5+ years), 246,000+ earnings call transcripts, 12,000+ companies across 70 countries, 170+ exchanges, and all 11 GICS sectors. Boolean and phrase queries let you write real screens: "pricing power" AND "elasticity", or "data center" NEAR capex, instead of hoping a single keyword lands.

Speaker tagging. A theme mentioned by an analyst in a question is not the same signal as a CFO committing to it in prepared remarks. The dataset carries 11M+ speaker segments, roughly 93% of which are attributed to a named speaker, tagged by role: executive, analyst, operator, attendee, shareholder. That means you can screen for what management said versus what the street pushed on — a distinction that changes how you read every hit.

Put those together and you can filter by ticker, sector, industry, country, and date range while searching the actual language of the call. That's the foundation every screening pattern below is built on.

Set up the MCP connector

If you already use Claude (or any MCP-capable client), wiring this up takes a minute.

1. Open your dashboard and go to Connectors → Generate.
2. Copy the connector URL, which looks like https://earningscalls.dev/u/mct_xxxx/mcp.
3. Add it as a custom MCP connector in your client and start prompting.

That's it — no local server, no API keys to juggle in your prompts. Full setup notes live on the MCP page and in the docs. If you want a deeper build, the walkthrough on how to build an earnings research agent with MCP and Claude picks up where this leaves off.

Screening patterns with example prompts

Here's where cross-market search earns its keep. Each pattern below is a repeatable prompt shape. Swap the theme, the sector, or the dates and you have a new screen.

Theme sweeps

The classic use case: pick a theme and see who's talking about it across the whole market, then rank by intensity.

Search all earnings calls from the last two quarters for mentions of "AI capex" and data center buildout. List the companies where executives (not analysts) raised or reaffirmed capital spending, with the exact quote and call date.

Screen the last 12 months of transcripts for "pricing power" and price increases. Group results by GICS sector so I can see which sectors are leaning on price vs. volume.

Find every call in the past 90 days where management mentioned layoffs, headcount reductions, or restructuring. Flag the ones where it's tied to margin guidance.

Because search is speaker-aware, that "executives, not analysts" instruction actually filters the results — you're not wading through analyst questions that merely name-check the theme.

Sector and industry filters

Themes rarely apply evenly. Constrain the universe first.

Within the Information Technology and Communication Services sectors, search for "generative AI" monetization and revenue contribution over the last three quarters. Which management teams put a number on it?

Limit to the Semiconductors industry. Find calls mentioning inventory correction or excess inventory in the last four quarters, and tell me whether the tone was improving or worsening quarter over quarter.

Date ranges and guidance changes

Timing is often the signal. Bound the window and look for the language of a change.

Search earnings calls between 2025-01-01 and 2025-06-30 for "raising our full-year guidance" or "raising the low end". Return company, quote, and date, sorted by date.

Compare mentions of "demand softening" or "weaker demand" in Consumer Discretionary calls this quarter versus the same quarter last year. Is the theme spreading or fading?

The date-range filter plus full-text phrasing is what turns a vague "how's demand looking" question into a datable, quotable trend line. For more prompt structures like these, the 10 Claude prompts for analyzing earnings calls via MCP post is a good companion.

Turning a screen into a shortlist

A theme sweep might return dozens of companies. The point isn't the long list — it's the shortlist you can actually act on. A few filters cut it down fast:

• Speaker role. Keep hits where an executive made the claim in prepared remarks or answered an analyst directly. Drop calls where the theme only appears in an analyst's question.
• Recency and repetition. A company mentioning a theme for three straight quarters is a different signal than a one-off. Ask the model to flag repeat mentions across a ticker's call history.
• Direction. "We're raising prices" and "we couldn't hold price" both match a pricing screen. Sort by sentiment or explicit direction so the shortlist reflects what you care about.
• Materiality. Prioritize calls where the theme is tied to guidance, margins, or a specific dollar figure — not passing color.

From the pricing-power screen above, keep only companies that (a) mentioned it in at least two of the last three calls and (b) tied it to gross margin expansion. Give me the final shortlist with one supporting quote each.

That single follow-up prompt takes you from a raw match list to a ranked set of candidates worth a real read.

Doing it at scale (500-ticker aggregates via the API)

MCP is ideal for interactive, conversational screening. When you want to run the same screen systematically across a defined universe — an index, a portfolio, a full coverage list — drop down to the REST API.

The API supports the same full-text and filter semantics, and it can sweep up to 500 tickers in a single call. That's the difference between a research question and a repeatable pipeline: point it at your 500-name universe, run a theme query, and get back every matching segment across all of them at once. Schedule it quarterly and you have an automated earnings-season monitor.

Typical scaled workflows:

• Universe monitors. Every quarter, sweep your coverage for a fixed basket of themes (capex, pricing, labor, guidance) and diff against last quarter's results.
• Aggregate counts. Track how many names in an index mention a theme over time — a simple, honest breadth indicator built from primary-source language.
• Feeding downstream systems. Pipe matched segments into a vector store for retrieval. If that's your direction, see how to feed earnings call transcripts into a RAG pipeline via MCP.

The docs cover the full-text query syntax, filter parameters, and the multi-ticker sweep endpoint in detail.

From screen to thesis

Screening is the funnel, not the finish line. The workflow that holds up looks like this:

1. Sweep the market for a theme with a full-text, speaker-aware query.
2. Filter to a shortlist using role, recency, direction, and materiality.
3. Read the handful of calls that survive — now you're reading with a question, not fishing.
4. Quote the primary-source language directly in your notes and models. Every hit comes with an attributed speaker and a date, so your thesis is grounded in what was actually said, not a paraphrase.
5. Re-run the screen next quarter to confirm the theme is building, not fading.

The compounding advantage is speed with rigor. You cover far more of the market than you could by reading, and every conclusion traces back to a named speaker on a dated call.

Get started

Screening earnings calls across companies is what separates a reactive research process from a proactive one. With MCP tools connected to EarningsCalls.dev, you can run theme sweeps, apply sector and date filters, build shortlists, and scale to 500-ticker aggregates — all from primary-source, speaker-tagged transcripts spanning 2020 to today.

Reading transcripts on the website is free. To unlock full-text screening via MCP and the API, Pro is $24.99/mo (5,000 requests/mo, 20/min) and Ultra is $39.99/mo (25,000 requests/mo, 60/min).

Pick a plan and start screening: see pricing.