> For AI agents: the documentation index is at https://platform.detectif.ai/llms.txt. Every page is also available as Markdown by adding `.md` to its path.

# MCP tools

The tools the DetectifAI MCP server provides, and their inputs.

The DetectifAI MCP server provides these tools. Your assistant chooses the tool and fills in the inputs from what you ask.

| Tool | What it does | Inputs |
|---|---|---|
| `delete_speaker` | Delete a speaker and its stored voice profile. | `speaker_id` |
| `detect_deepfake` | Check a recording (longer than 4 s, up to 120 s) for AI-generated speech. | `audio_base64`, `filename`, `idempotency_key` (optional) |
| `enroll_speaker` | Add one enrollment sample for a speaker (creates the speaker on first use) with Eidos. | `speaker_id`, `audio_base64`, `filename`, `display_name` (optional), `idempotency_key` (optional) |
| `get_credit_balance` | Return the account's available and reserved credits. | none |
| `get_operation` | Fetch a stored operation: status and, for 30 days, its result. | `operation_id` |
| `list_speakers` | List enrolled speakers in this account, newest first. | `limit` (optional), `cursor` (optional) |
| `verify_speaker` | Compare a recording (3-60 s, at least 2 s of speech) with an enrolled speaker using Eidos. | `speaker_id`, `audio_base64`, `filename`, `detect_deepfake` (optional), `idempotency_key` (optional) |
Audio tools take the recording as base64 text in `audio_base64` (up to 25 MB decoded) plus its `filename`. Pass an `idempotency_key` and reuse it when retrying, so a retry returns the stored result instead of running again. See [Passing audio to tools](https://platform.detectif.ai/docs/mcp/#passing-audio-to-tools).