# The hosted MCP server

Intervieux runs a hosted remote MCP server at app.intervieux.ai/mcp that an MCP client like Claude or ChatGPT connects to directly, completing a real OAuth 2.1 and PKCE authorization flow with dynamic client registration and a genuine consent screen, exposing around 40 tools with nothing to install and no local process to run.

Most MCP servers still assume a local process, a config file with a filesystem path, and an API key pasted somewhere on disk. A hosted server removes all three: point a client at one URL, log in through a browser, and the tools are available.

## What it does

The hosted server exposes the same tool set as Intervieux's standalone MCP package, roughly 40 tools spanning jobs, applications, pipeline, scheduling, documents, and talent search, but reached over HTTP instead of a local stdio process. It runs in stateless mode, meaning it doesn't hold session state between requests, since it's a multi-tenant server handling many employers' credentials at once. Getting connected means going through a real OAuth 2.1 authorization flow with PKCE, ending at Intervieux's own consent screen, the same dashboard an employer already logs into, rather than a fake or simplified approval page.

## How it works

1. **Point an MCP client at the server** — An MCP client, Claude or ChatGPT's connector interface among them, is configured with the single URL app.intervieux.ai/mcp. No local install, no path on disk.
2. **The client registers itself** — Dynamic Client Registration lets the connecting client register automatically the first time it connects, rather than requiring a client ID to be manually created and copied over ahead of time.
3. **Authorization runs through OAuth 2.1 with PKCE** — The client is sent through a standard browser-based OAuth flow with PKCE, landing on Intervieux's real login and then its consent screen, the same one wrapping every other authenticated employer page in the dashboard.
4. **Consent is explicit** — The employer sees and approves what the connecting client is asking for on that consent screen before any access is granted, rather than access being implied by simply having a key.
5. **Every tool call authenticates and scopes independently** — Because the server is stateless, each request carries its own credential. That credential forwards internally to the same REST API scope checks the ATS API itself uses, so a tool call missing the right scope comes back as a clean MCP error instead of a raw HTTP failure.

## One codebase behind both surfaces

The hosted MCP server doesn't maintain its own separate implementation of each tool. It reuses the exact same tool definitions the standalone MCP package ships, so a tool behaves identically whether it's reached through the hosted server or the local package, and a fix or addition only has to happen once.

## Real OAuth, not a shortcut

Dynamic Client Registration, PKCE, and a genuine consent screen are the full OAuth 2.1 flow, not a simplified stand-in. Discovery metadata is published at the standard well-known endpoints, so a compliant client can find everything it needs to connect without being told the exact URLs by hand.

## Stateless by necessity

A single hosted server handles requests from many different employers, so it can't hold onto session state between calls the way a single-user local server might. Every request re-establishes who it's acting for from the credential it carries, which is also what makes the server safe to run as one shared multi-tenant process instead of one per employer.

## Access can be seen and revoked

Once a client is connected, it shows up as a connected app in the Developer / API screen, where an owner or admin can revoke that grant directly, cutting off that client's access without touching anything else.

## Who this is for

This is for connecting an AI assistant like Claude or ChatGPT straight to a company's Intervieux hiring data without installing anything or managing a local server process, and for any MCP-compatible tool that would rather authenticate through a real browser login and consent flow than a copy-pasted API key.

## Frequently asked questions

### Do I need to install anything to use the MCP server?

No. The server is hosted at app.intervieux.ai/mcp. An MCP client connects to that URL directly, there's no local process, package install, or filesystem path involved.

### How does authentication work?

Through a standard OAuth 2.1 flow with PKCE and Dynamic Client Registration, ending on Intervieux's real consent screen, the same dashboard login and approval flow used elsewhere in the app, not a simplified stand-in.

### How many tools does the MCP server expose?

Around 40 tools, covering jobs, applications, pipeline, scheduling, documents, and talent search, the same tool set the standalone MCP package registers.

### Can I see or revoke which clients have connected?

Yes. Connected OAuth and MCP clients show up in the Developer / API screen's connected apps section, where an owner or admin can revoke a client's access directly.

### Does it work with Claude and ChatGPT?

Yes, it's built as a standard MCP server reachable by any compatible MCP client, including Claude and ChatGPT connectors, over the hosted URL with no local setup required.

## Related pages

- [The ATS REST API](/features/ats-api)
- [Google and Microsoft connected accounts](/features/connected-accounts)
- [Team members and roles](/features/team-roles)
- [Frequently asked questions](/faq)

## Connect an AI assistant to your hiring data in one URL

Point Claude or ChatGPT at app.intervieux.ai/mcp, sign in, and approve access through a real consent screen.

Start practicing free: https://www.intervieux.ai/register · Hire with Intervieux: https://www.intervieux.ai/employers/signup