AI Interview Agents MCP Server
# AI Interview Agents — MCP Server
[](https://glama.ai/mcp/servers/dpkdhingra91/aiia-mcp-server)
The official [Model Context Protocol](https://modelcontextprotocol.io) server for [AI Interview Agents](https://www.aiinterviewagents.com), a voice-AI hiring platform. It lets any MCP client (Claude, Cursor, custom agents) drive the recruiting workflow in natural language: create roles, screen CVs, schedule and manage candidate interviews, and read back scored reports with transcripts.
The live voice interview itself is out of scope here (that runs on a separate real-time service). This server *triggers* interviews and *reads back* the results.
## What it exposes
- **Roles** — list your roles, or create one from a job description.
- **Candidates & CVs** — parse dropped CV/PDF files, normalize messy pasted candidate lists (names, emails, phone numbers, experience, skills), and reassign candidates between roles.
- **Interviews** — schedule interviews (with a `dry_run` mode that normalizes and previews without sending anything or consuming quota), reschedule, cancel, and send reminders.
- **Reports** — read the candidate pipeline and pull back scored interview reports with transcripts.
## Use the hosted server (recommended)
The server runs remotely with OAuth sign-in. Point any Streamable-HTTP MCP client at:
```json
{
"mcpServers": {
"ai-interview-agents": {
"type": "streamable-http",
"url": "https://mcp.aiinterviewagents.com/mcp"
}
}
}
```
Your client discovers the auth flow automatically via `/.well-known/oauth-protected-resource` and signs you in with your AI Interview Agents account. It is also listed in the [official MCP Registry](https://registry.modelcontextprotocol.io) as `com.aiinterviewagents/interviews`.
## Run it locally (stdio)
```bash
pip install -e .
export AIIA_FIREBASE_TOKEN="<your token>" # or: aiia-mcp login
aiia-mcp
```
`AIIA_BACKEND_URL` overrides the API base if you are pointing at your own deployment.
## How auth works
The server holds no account secrets. Each request carries the user's own Firebase ID token, forwarded to the backend as a bearer token — the same trust path as the web app. For the hosted transport, an OAuth layer (dynamic client registration + PKCE) bridges MCP clients to Firebase sign-in.
## Development
```bash
pip install -e ".[dev]"
pytest
ruff check .
```
## License
MIT — see [LICENSE](LICENSE).
TDQS
Scored across 18 tools
Tools have distinct purposes with detailed descriptions, but there is slight overlap between schedule_interview and schedule_screened_candidates (both schedule interviews, one for unscreened and one for screened candidates). Overall, the differences are clear from the descriptions.
All tool names follow a consistent verb_noun pattern in snake_case (e.g., list_meetings, create_screening_role, run_screening). No mixing of conventions like camelCase or irregular verbs.
With 18 tools, the set covers the full interview management lifecycle without being overwhelming. Each tool serves a specific function, and the count is appropriate for the domain.
The tool surface covers the screening pipeline (from role creation to scheduling screened candidates) and interview management (schedule, cancel, reschedule, list, get details, report, reminders). Minor gaps include no tool for updating candidate email or deleting roles, but core workflows are well-supported.