Predict how real human brains respond to your content.

Authenticate

Sign in at neuro-agi.com → Settings → API, then Mint key. The secret (sk_live_…) is shown once — copy it then. Send it on every request:

Authorization: Bearer sk_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Add NeuroAGI to your coding agent

One command wires the MCP server into any agent — Claude Code, Codex, Cursor, Windsurf, anything that speaks MCP:

claude mcp add neuroagi --env NEUROAGI_API_KEY=sk_live_… -- npx -y @neuroagi/mcp

No key? It still runs. Install without the --env flag and your agent pulls a real, precomputed demo scan (the keyless demo_scan tool) so you can see what NeuroAGI produces before signing up:

claude mcp add neuroagi -- npx -y @neuroagi/mcp

Packages: @neuroagi/mcp (the server) and @neuroagi/cli (the same calls in your terminal).Building for an agent? See the agent reference →

Submit a scan

Send the content you want scored — exactly one of a URL or text.

{
  "input": { "type": "url", "url": "https://yourcdn.com/ad.mp4" },
  "model": "qualia",
  "lenses": ["attention", "purchase_intent", "manipulation"],
  "options": { "include_reasons": true, "include_benchmark": true },
  "webhook_url": "https://yourapp.com/hooks/neuroagi"
}

The scan queues and returns an id immediately:

{ "scan_id": "qualia_run_abc123", "status": "queued", "lenses": ["attention", "purchase_intent", "manipulation"] }

Get a scan

While in flight you get { scan_id, status }. When status is complete you get the full rich shape (below). status ∈ queued, processing, complete, error. A scan that isn't yours returns 404. If you passed webhook_url, we POST the same completed payload there once (best-effort; polling is the fallback).

The result shape

One shape, always — a headline score + grade, a per-second timeline, and (with include_reasons) moments + a summary.

{
  "scan_id": "qualia_run_abc123",
  "status": "complete",
  "duration_sec": 30,
  "lenses": ["attention", "purchase_intent", "manipulation"],
  "score": 78,                                                      // alias of overall_score
  "grade": "B",
  "meta": { "scales": { "raw": "z-scored; 0 = baseline, +1 ≈ 1 SD above", … } },
  "timeline": [
    {
      "second": 0,
      "scores": { "attention": 72, "purchase_intent": 61, "manipulation": 47 },
      "raw": { "Visual": 0.13, "Limbic": 0.14, "…": 0.0 },          // options.include_raw
      "raw_labeled": { "Visual": 0.13, "Emotional": 0.14, "…": 0.0 }, // founder vocabulary
      "benchmark": { "attention": { "percentile": 84, "label": "top" } }, // include_benchmark
      "reasons": { "attention": "…" }                              // include_reasons
    }
  ],
  "network_profile": [ { "network": "Limbic", "label": "Emotional",  // include_raw — clip rollup
      "avg": 0.57, "peak_sec": 12, "read": "the emotional pull is rising" }, … ],
  "moments": [ { "second": 15, "label": "peak", "quote": "…", "why": "…" } ], // include_reasons
  "buy_moments": [ { "peak_sec": 20, "strength_pct": 78, "quote": "…",  // include_reasons
      "insight": "purchase intent peaks here" } ],
  "summary": "…",                                                   // include_reasons
  "fmri_url": "https://…signed…"                                    // include_fmri
}

• score / grade — the headline 0–100 brain-engagement prediction + its letter grade. (overall_score is the same value.)
• timeline — one entry per second. scores is a 0–100 read per lens (50 = neutral). raw, raw_labeled, benchmark, reasons appear only with their option.
• network_profile — a clip-level rollup (with include_raw): each network's average + a plain read. meta.scales explains every unit.
• moments + summary — the notable seconds and a plain-English read. Need include_reasons.
• fmri_url — signed link to the fMRI artifact, only with include_fmri.

The lenses

Three 0–100 per-second reads. Pass any subset in lenses; omit to get all three.

Purchase intent → buy moments. With include_reasons, the scan also returns buy_moments — the timestamped windows where purchase intent peaks, each with the peak second, a strength, the quote said there, and a plain-language insight. It's the "when does this make someone want to buy" layer on top of the per-second purchase_intent curve.

Options & raw export

Raw / parcellation. include_raw exposes the model's underlying signal — the 7 Yeo networks (Visual, Somatomotor, Dorsal Attention, Ventral Attention, Limbic, Frontoparietal, Default) at each second. Off for the compact lens read; on for the neural substrate.

Reading the raw output

The raw block is the 7 brain networks as z-scores: 0 = population baseline, +1 ≈ one standard deviation above, negative = below. Positive means that network is more engaged than average. Here's what each network is and what a rise vs. a fall means — the full machine-readable version is GET /api/v1/parcellation.

Every scan also returns a clip-level network_profile (each network's average across the whole scan + a plain read), a meta.scales legend, and — for interpreting raw values programmatically — the public GET /api/v1/parcellation reference:

curl https://www.neuro-agi.com/api/v1/parcellation
# → { "object": "parcellation", "scale": {…}, "networks": [ { "network": "DorsalAttention",
#     "label": "Attention", "name": "focus network", "rises_when": "…", "falls_when": "…" }, … ],
#     "lenses": [ { "lens": "attention", "driven_by": [ … ] }, … ] }

Other endpoints

GET /api/v1/demo is the only endpoint that needs no auth:

curl https://www.neuro-agi.com/api/v1/demo
# → { "object": "demo_list", "count": 1, "demos": [ { "score": …, "timeline": […] } ] }

Pricing

Scans run on Qualia, the active model (the visual specialist — video, visual + audio). Per-scan price drops automatically as monthly volume grows; no plan to pick.

• Credits never expire. Add them in Settings → API.
• Any signed-up user can create a key. API credits are separate from a NeuroAGI subscription.

Quick start (curl)

# Submit
curl https://www.neuro-agi.com/api/v1/scans \
  -H "Authorization: Bearer sk_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{ "input": { "type": "url", "url": "https://yourcdn.com/ad.mp4" } }'
# → { "scan_id": "qualia_run_abc123", "status": "queued" }

# Poll
curl https://www.neuro-agi.com/api/v1/scans/qualia_run_abc123 \
  -H "Authorization: Bearer sk_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
# → { "scan_id": "qualia_run_abc123", "status": "complete", "score": 78, … }

Errors

Standard HTTP status codes. Bodies look like:

{ "error": "unauthorized", "message": "Missing, invalid, or revoked API key." }