Earnings calls are one of the richest, least-structured sources of information in public markets. Every quarter, thousands of management teams read prepared remarks, then face unscripted questions from analysts. What they say — and how they say it — moves stocks, reshapes forecasts, and signals shifts long before they show up in the numbers. The problem has never been that this information is secret. It is public. The problem is getting it in a form a machine can actually work with, at scale, across every company you care about.
This is the complete guide to doing exactly that in 2026. It covers the two ways to access earnings call transcripts programmatically — a traditional REST API and the newer Model Context Protocol (MCP) connector — what good coverage looks like, how to set each one up, and the concrete use cases they unlock. Whether you are a quant building a screener, a developer wiring transcripts into a Retrieval-Augmented Generation (RAG) pipeline, or an analyst who just wants Claude to read the last call for you, this page is the hub. It links out to focused deep-dive guides for each scenario.
What are earnings call transcripts (and why programmatic access matters)
An earnings call is the quarterly (sometimes semi-annual) conference call a public company hosts to discuss results. A transcript is the verbatim record of that call: the CEO's and CFO's prepared remarks, the operator's instructions, and the question-and-answer exchange between executives and sell-side analysts. The Q&A section is often where the real signal lives, because it is unscripted.
Reading a single transcript is easy — you can do it free on any transcript site, including earningscalls.dev with no signup. The challenge is scale and structure. If you want to compare margin commentary across 40 companies, track how often "macro headwinds" appears in a sector over eight quarters, or feed the latest calls into an AI agent every morning, copy-pasting from web pages does not work. You need:
- Structured data, not HTML — each transcript broken into speaker segments with names, roles, and timestamps.
- Breadth, so you are not limited to a handful of US mega-caps.
- History, so you can measure change over time.
- Freshness, so a call that ended twenty minutes ago is already queryable.
Programmatic access — via an API or an MCP connector — is what turns a pile of transcripts into a dataset you can query, screen, and reason over.
Two access models: REST API vs MCP connector
There are two fundamentally different ways to consume earnings call transcripts programmatically, and they serve different jobs.
A REST API is the classic integration. Your code makes HTTP requests, authenticates with an API key, and receives JSON. It is deterministic, scriptable, and ideal for backends, cron jobs, dashboards, and data pipelines. You control exactly what is fetched and when.
An MCP connector is newer. The Model Context Protocol is an open standard that lets AI assistants like Claude call external tools directly. Instead of writing request code, you paste one connector URL into your AI client, and the model gains a set of tools it can invoke on your behalf — "get the latest call for NVDA," "search transcripts for supply-chain commentary," "list this quarter's tech calls." The AI decides which tool to call based on your natural-language question.
Here is how they compare:
| Dimension | REST API | MCP connector |
|---|---|---|
| Caller | Your code | An AI assistant (Claude, Cursor, etc.) |
| Interface | HTTP + JSON | Tools invoked by the model |
| Auth | X-API-Key header |
Token embedded in connector URL |
| Best for | Pipelines, backends, dashboards, batch jobs | Interactive research, agents, ad-hoc Q&A |
| Determinism | Full control over every request | Model chooses which tools to call |
| Setup | Write and maintain client code | Paste one URL into your AI client |
| Learning curve | Standard REST knowledge | None — natural language |
They are not mutually exclusive. Many teams use the REST API to power scheduled ingestion and the MCP connector to let analysts explore the same data conversationally. For a deeper side-by-side, see MCP vs REST API for earnings call data. If the concept of an MCP server is new to you, start with What is an earnings call MCP server?.
What good coverage looks like (depth, breadth, freshness, speaker structure)
Not all transcript sources are equal. The value of a provider is defined by four things.
Breadth. How many companies, and where? A source limited to the S&P 500 is useless the moment you care about a European industrial or an Asian semiconductor supplier. The earningscalls.dev dataset spans 12,000+ companies across 70 countries, 170+ exchanges, and all 11 GICS sectors. That is genuinely global coverage, not just US large-caps.
Depth (history). A single quarter tells you almost nothing. Trends, seasonality, and changes in tone only emerge over years. Coverage here runs from 2020 to the present — more than five years — totaling 246,000+ earnings call transcripts.
Speaker structure. This is what separates a real dataset from a wall of text. Every transcript is broken into speaker segments — 11M+ of them — each classified by role: executive, analyst, operator, attendee, or shareholder. About 93% of segments have a named speaker, so you can filter to "everything the CFO said" or "all analyst questions" instead of parsing paragraphs by hand. This structure is what makes serious screening and RAG possible.
Freshness. Stale data is worthless for anything real-time. New calls land within minutes of the call ending, and the API exposes cursor-based polling so you can efficiently pull only what is new since your last check.
When you evaluate any provider, hold it to these four bars. Depth without breadth traps you in one market. Breadth without speaker structure gives you text you still have to parse. Freshness without history means no trend analysis.
How to access transcripts with the REST API
The REST API is the foundation for any programmatic workflow.
- Base URL:
https://earningscalls.dev/api/v1 - Authentication: send your key in the
X-API-Keyheader on every request. - OpenAPI spec: the full machine-readable spec lives at
https://earningscalls.dev/openapi.json, so you can generate a typed client in any language.
Core endpoints include:
GET /earnings— list and filter earnings calls (by ticker, date, sector, and more).GET /earnings/latest— the most recent calls, ideal for polling.GET /search— full-text search across transcripts and speaker segments.GET /transcripts/{id}— fetch a full transcript with its structured speaker segments.
A minimal request looks like this:
curl -H "X-API-Key: your_key_here" \
"https://earningscalls.dev/api/v1/earnings/latest"
Two features matter at scale. Cursor-based polling lets you fetch only calls added since your last request, which is how you keep an ingestion job cheap and idempotent. And aggregate endpoints accept up to 500 tickers, so you can pull a whole watchlist or index constituents in one call instead of hammering the API ticker by ticker.
Full request/response details, parameters, and examples are in the API documentation. The spec is versioned under /api/v1, so integrations stay stable.
How to access via MCP
If you work inside an AI assistant, the MCP connector is dramatically faster to set up than writing API code.
To generate your connector, go to the Dashboard → Connectors → Generate. You will get a personal URL of the form:
https://earningscalls.dev/u/mct_xxxx/mcp
Your access token is embedded in that URL, so treat it like a secret. Then add it to your AI client:
- Claude Desktop — add the connector under Settings → Connectors.
- claude.ai (web) — add it as a custom connector in your account settings.
- Claude Code — register the MCP server via the CLI.
- Cursor — add it to your MCP configuration.
Once connected, the model gains roughly 13 tools covering the same ground as the REST API: fetching the latest call for a ticker, listing calls by company, searching transcripts, pulling speaker segments, browsing sectors and industries, checking upcoming earnings, and more. You simply ask in plain language — "summarize Apple's most recent earnings call and pull out any guidance changes" — and the model chooses and chains the right tools.
For a step-by-step walkthrough with screenshots, see How to connect earnings call transcripts to Claude with MCP, and browse the MCP overview page for the current tool list.
Common use cases
The same dataset powers very different workflows. Here are the main ones, each with a focused guide.
Fundamental research. Read, summarize, and cross-reference calls without leaving your AI assistant. Ask for guidance changes, pull every mention of a product line, or compare tone across quarters. A good starting point is 10 Claude prompts for analyzing earnings calls via MCP — copy-paste prompts that work out of the box.
Market-wide screening. Instead of reading one call, screen thousands. Find every company in a sector that raised guidance, or flag mentions of a specific supplier across the market. The speaker structure and 500-ticker aggregates make this practical. See Screen earnings calls across the market with MCP tools.
RAG pipelines. Feed transcripts into a vector store and build a grounded question-answering system over years of calls. The pre-segmented speaker data chunks cleanly, which improves retrieval quality. Full walkthrough: Feed earnings call transcripts into a RAG pipeline via MCP.
Autonomous agents. Build an agent that researches a company end to end — pulling the latest call, comparing it to prior quarters, and writing up the delta. See Build an earnings research agent with MCP and Claude.
Real-time alerts. Because new calls land within minutes and polling is cursor-based, you can build an agent that watches for fresh calls and alerts you when a company you follow reports — or when specific keywords appear. See Real-time earnings call alerts with an MCP agent.
Choosing a provider / what to look for
When you compare earnings transcript providers, weigh these criteria:
- Global breadth — does coverage extend beyond US large-caps to the exchanges and countries you actually track?
- History — is there enough back-catalog to measure change over time?
- Structured speaker segments — can you filter by role and named speaker, or is it undifferentiated text?
- Freshness and polling — how fast do new calls appear, and can you fetch only what changed?
- Both access models — is there a REST API and an MCP connector, so you are not locked into one workflow?
- A clean spec — an OpenAPI definition means you can generate clients and integrate quickly.
- Sensible limits — batch/aggregate endpoints so you are not rate-limited into thousands of tiny requests.
earningscalls.dev is built specifically around this checklist: five-plus years of history, 70 countries, 11M+ role-classified speaker segments, minute-level freshness, cursor polling, 500-ticker aggregates, a published OpenAPI spec, and both a REST API and an MCP connector.
Getting started
Reading transcripts on the website is completely free with no signup — go browse a few calls to see the data first-hand. Programmatic access via the API or MCP is a paid product:
| Plan | Price | Requests | Rate limit | Notes |
|---|---|---|---|---|
| Pro | $24.99/mo | 5,000/mo | 20/min | For individuals and small projects |
| Ultra | $39.99/mo | 25,000/mo | 60/min | For heavier research and agents |
| Enterprise | Custom | 100,000/mo | — | Includes webhooks |
To begin:
- Read a few transcripts free at earningscalls.dev to confirm the coverage fits.
- Pick a plan on the pricing page.
- In the Dashboard, grab your API key (for REST) or generate an MCP connector URL.
- Follow the docs for the API, or the MCP guide for the connector.
FAQ
Do I need to pay just to read a transcript? No. Reading transcripts on the website is free and requires no signup. Payment is only for programmatic access through the API or MCP connector.
Should I use the REST API or the MCP connector? Use the REST API for backends, pipelines, and scheduled jobs where you want deterministic control. Use the MCP connector when you want an AI assistant to research and reason over transcripts conversationally. Many teams use both. See MCP vs REST API for earnings call data.
How fresh is the data? New calls are typically available within minutes of the call ending. Cursor-based polling lets you fetch only what is new since your last request.
What does coverage include? More than 246,000 transcripts from 12,000+ companies across 70 countries, 170+ exchanges, and all 11 GICS sectors, going back to 2020 — with 11M+ speaker segments classified by role.
Get started
The fastest path from "I want earnings call data" to a working integration is short: read a few transcripts free, choose a plan, and generate your key or connector.
Pick a plan on the pricing page, then head to your Dashboard to get your API key or generate an MCP connector URL. In minutes you will have every earnings call — five years of history, 70 countries, structured by speaker — available to your code or your AI assistant.