Platform
MCP & Skills for LLMs
Bring ScamAI detection to AI agents — an MCP server and agent skills that let LLMs verify media and documents mid-task.
- An MCP server any LLM or agent can call
- Verify selfies, documents, and video mid-task
- Packaged skills for agent frameworks
What it does
An MCP (Model Context Protocol) server that exposes ScamAI detection as tools any compatible LLM or agent can call — verify a selfie, score a document, check a video — plus packaged skills that wrap full verification workflows for agent frameworks.
Detection your agents can trust
Agents get the same scored results and evidence as the API — so an automated decision is backed by the exact signals a human reviewer would see, not a guess from the model.
Tool Calls: how the MCP works
ScamAI's MCP server exposes detection as typed tools any compatible agent can call mid-task — check a selfie, score a document, verify a video — backed by the same Eva V1.6 engine as the REST API.
Putting verification inside agent workflows
Give your agents a verification step so they check media before acting on it, and escalate anything risky to a human.
/ROI
What this is worth
- One AI engine for every surface — add modules without adding vendors, contracts, or review queues.
- Evidence-backed results drop straight into your decisioning: auto-decline, step-up, or route to review.
- Live in a day: one REST API call in, a scored result out — no model training, no data-science team required.
Common questions
What is an MCP server for deepfake detection?
MCP — the Model Context Protocol — is an open standard that lets LLMs and agents call real tools instead of guessing. ScamAI's MCP server exposes detection as tools: an agent submits a selfie, document, image, or video mid-task and receives a probabilistic confidence score plus the evidence signals that fired, back in its context. The tools are backed by the same Eva V1.6 engine as the REST API, so agent-initiated checks carry the same weight as any other result.
How can AI agents verify images and documents?
Through tool calls mid-task. An agent reviewing a loan application calls a verification tool for each attached document; the media is scored by the detection backend and the result — score plus named signals — returns into the agent's context, where it can branch: proceed, request better documentation, or escalate to a human queue. Encode escalation rules in the skill rather than the prompt, so high-risk results always trigger a handoff no matter how the conversation has drifted.
Which LLMs and agent frameworks work with ScamAI's MCP server?
Any runtime that speaks the Model Context Protocol. MCP is an open standard with broad adoption across major assistants and agent frameworks, so integration is configuration rather than custom development: point your agent runtime at the ScamAI server and the detection tools become available to the model. For teams building on frameworks with their own tool conventions, ScamAI also packages skills that wrap complete verification workflows, and the underlying REST API remains available for anything the protocol does not cover.
Building an agent that needs detection?
15 minutes, on your own media. Pick a slot and leave with a result.