Connect an assistant through MCP
Exterview uses the Model Context Protocol (MCP) to connect authorized assistants to hiring workflows. Ask your Exterview representative for the endpoint and supported setup for your workspace. After signing in, use tool discovery to see which operations your connection can call.- Connected Agent Access covers setup for Claude and ChatGPT.
- MCP protocol and troubleshooting covers OAuth, tool discovery, requests, and errors.
- Tool catalog documents the planned read-only directory release. Check your connection’s tool list for current availability.
Upcoming: the partner API
How the partner API is shaped
- GraphQL-first, one endpoint. A single endpoint serves every intelligence capability: create the entities, trigger an evaluation, then read the result. A small REST surface exists only for a health check and typed bulk-file ingest (candidate and job uploads). Everything else is GraphQL.
- Scoped to your organization. Every request runs inside your organization. Your organization and role are derived from your credentials on the server, never trusted from the request body. There is no access to another customer’s data.
- Poll-based, no outbound webhooks. Exterview does not send outbound webhooks to partner integrations today. Long-running operations, like an AI phone screen, are read by polling the relevant record until its status is ready, rather than waiting for a push notification.
- Filtered and paginated reads. List operations accept arguments to narrow results (by status, by parent entity, by search term) and are paginated rather than returned in one call.
A typical integration flow
Regardless of which evaluation you’re driving, the shape is the same:1
Create or locate the entities
A job and a candidate must exist first, created individually or in bulk.
2
Trigger the evaluation
Call the capability’s trigger operation. This starts async work; it doesn’t
return the finished result.
3
Poll for completion
There’s no outbound webhook. Poll the record until its status reaches a
terminal state.
4
Read the AI output
Once ready, read the report: the structured, cited evaluation.
5
Read the stack rank
See where the candidate now sits, rolled up across every completed round for
the job.
Partner authentication
Partner access will use API-key or OAuth client-credentials with scopes (read, write, and higher-sensitivity scopes for actions that spend money, message candidates, or change a hiring decision). This is Upcoming. See Authentication for the full scope model.
Conventions
- Errors follow the standard GraphQL
{ data, errors }shape; a custom error-code taxonomy isn’t published yet. - Rate limits and idempotency for the partner API are Upcoming: per-key limits, quota headers, and required idempotency keys on money- or comms-impacting mutations.
- Pagination will use Relay-style (
edges/pageInfo) for GraphQL connections and REST-style (perPage/pageNumber) for a few endpoints like the stack-rank read.
FAQs
Can I call the API today?
Can I call the API today?
For MCP access, see Connected Agent Access. Partner GraphQL credentials are upcoming. The public Career Page is open.
Is it REST or GraphQL?
Is it REST or GraphQL?
GraphQL-first, with a small REST surface for bulk file ingest.
How is my data isolated from other customers?
How is my data isolated from other customers?
Every request is scoped to your organization, derived from your credentials on
the server. There is no access to another customer’s data.
Are outbound webhooks available?
Are outbound webhooks available?
Not today. Long-running operations are read by polling the relevant record until it’s ready.
Where's the full technical reference: exact fields, types, and operation names?
Where's the full technical reference: exact fields, types, and operation names?
See the MCP tool catalog for the planned directory release. The partner GraphQL reference will accompany partner API access.
Related
- Authentication. The scope model and the prerequisite that gates the API.
- Capabilities. What each capability does and why you’d use it.

