The entire value proposition of the Model Context Protocol is that it's a standard. One server, one URL - and every client that implements the protocol can use it. No per-app integrations, no SDKs to install, no rebuilding the same bridge five times.
EarningsAPI leans on that fully. The hosted MCP server at mcp.earningsapi.io sits behind a single personal connector URL, and that identical URL works in Claude Desktop, claude.ai in the browser, Claude Code in the terminal, Cursor in the editor, and any other MCP-capable client. Learn the connector once, use it everywhere.
This guide covers the universal pattern first, then the per-client steps. All roads lead to the same place: an assistant that can search, browse, and read across 250,945 earnings call transcripts from 12,728 companies, with history back to 2020.
Step zero: your connector URL
Everything starts with one credential - the connector URL. The access token lives inside the URL itself, which means no API-key field, no OAuth flow, no environment variables.
To get it:
- Sign in and open your dashboard.
- Navigate to Connectors.
- Hit Generate.
The result looks like:
https://earningsapi.io/u/mct_xxx/mcp
The transport is streamable HTTP, and the mct_xxx segment is your token - so the URL is a secret. Whoever holds it can spend your quota. If it ever leaks, revoke it on the Connectors page and mint a new one.
Each client below is just a different place to paste this URL.
Claude Desktop
Custom connectors are built into the settings, so no config-file editing:
- Settings
- Connectors
- Add custom connector
- Paste the URL.
Open a fresh conversation and try something like "How did semiconductor companies talk about pricing on their most recent calls?" - Claude discovers the tools and starts calling them.
claude.ai in the browser
Identical flow to the desktop app:
- Settings
- Connectors
- Add custom connector
- Paste the URL.
Streamable HTTP runs fine in the browser, and the connector is attached to your account rather than your device - configure it once and it's live on web and desktop simultaneously.
Claude Code
In the terminal client, adding the server is a one-liner:
claude mcp add --transport http earningsapi "https://earningsapi.io/u/mct_xxx/mcp"
Substitute your real URL. The --transport http flag selects streamable HTTP, which is what the connector speaks. From then on, transcript search and retrieval happen inline while you work - handy if you're scripting analysis, filling a research notebook, or composing a larger agent pipeline from the shell.
Cursor
Cursor's agent can use the same server, putting earnings data next to your code:
- Open Settings.
- Locate the MCP section.
- Add a server with the HTTP / streamable transport.
- Enter your connector URL.
After Cursor registers the server, the editor agent can fetch a company's latest call or run a corpus-wide theme search without you leaving the project.
Every other MCP client
The protocol is the contract, so the recipe never changes:
- Add a new MCP server in the client's settings.
- Select the streamable HTTP transport - the connector doesn't offer stdio or SSE.
- Set the server URL to your connector URL.
That covers editors, chat frontends, and custom agents built on any MCP framework. Since auth travels in the URL, extra credential fields usually stay empty. If the client asks you to name the server, use earningsapi so it's recognizable in tool lists.
One caveat: a client that only supports local stdio servers can't reach a remote connector. For those environments, use the REST API directly (more below).
What your assistant can do once connected
The connector hands your assistant a toolset for working with earnings calls:
- Corpus-wide search - locate every mention of a phrase or theme ("AI capex", "inventory correction") without knowing which company said it.
- Browsing and filtering - companies and calls by ticker, sector, industry, or region.
- Full transcript retrieval - any call, including the newest one for a ticker.
- Speaker segments - isolate what one executive said, or step through the Q&A, instead of scanning a wall of text. All 11.9M segments are role-tagged.
- Calendar and stats - recent calls, upcoming earnings dates, coverage numbers.
Behind those tools is production coverage, not a sample set: 250,945 transcripts, 12,728 companies, every GICS sector, role-tagged segments, history to 2020. That depth is why one prompt can honestly answer cross-company and cross-sector questions.
The complete tool list is on the MCP page; the full reference is in the docs.
MCP vs. REST: pick by who's asking
Choose MCP when a language model drives the queries - exploratory research, chained lookups, "let the assistant figure out what it needs." The connector is the shortest path and works identically across every client above.
Choose the REST API when your own code drives - dashboards, cron jobs, ETL, anything deterministic. The base URL is:
https://earningsapi.io/api/v1
An OpenAPI spec is available for generating typed clients, and REST is also your fallback for stdio-only MCP clients.
Both interfaces read the same corpus and draw from the same quota, so you can mix them freely. Plans are billed quarterly - Basic at $105, Pro at $145, Ultra at $515 - with the request allowances listed on the pricing page.
Two minutes to connected
Generate the URL, paste it into your client of choice, ask a question. Because MCP is a standard, that one URL follows you from Claude Desktop to Cursor to Claude Code to whatever tool ships next - no re-integration, ever.