MCP

RoastIQ, right inside Claude.

Ask Claude to score an ad. RoastIQ runs in the background and the verdict, five KPIs, benchmark position, comes back in the same conversation. Same account, same benchmark pool as the web app.

Get your API key →Read the FAQ

One-time setup

Three steps. No new account.

The SaliencyLab MCP server exposes RoastIQ and BuyerLens as tools over Streamable HTTP. Claude calls it directly with a scoped API key, no OAuth dance.

Create a live API key

In your workspace, go to Settings → API keys and generate a key (starts with sl_live_). It inherits your role and workspace.

Add the server in Claude Code

Run the command on the right once. Claude Desktop's remote-MCP support (Settings → Connectors) works the same way with a custom header.

Ask Claude to score an ad

Attach the file or point to it, and Claude routes the call to RoastIQ automatically.

terminal

claude mcp add --transport http saliencylab \
  https://ecyudrlpprqfxccvqkyd.supabase.co/functions/v1/mcp-server \
  --header "Authorization: Bearer sl_live_YOUR_KEY"

The workflow

Stay in the chat. Ship the fix.

You ask, in plain language

Drop the ad file into the conversation and ask Claude to score it.

“Score this ad against our DTC skincare benchmark before we brief the edit.”

Claude calls RoastIQ

The tool call routes to the same pipeline as the web app — no separate account, no re-upload.

↳ calling score_creative(asset_id, project_id, category="dtc-skincare")

The verdict returns inline

Five KPIs, composite, and the single next move, rendered right in the chat. Video scores arrive via a status poll, same as the web app.

Coca-Cola · Holidays 2025SHARPEN

81Skip

79Notice

82Brand

63Sell

81Build

Ship or iterate, same thread

Ask Claude to run BuyerLens on the same result, or compare it against a previous cut, still without leaving the chat.

Tool surface

13 tools. One dispatch layer.

Every call runs through the same auth, quota, and workspace-tenancy logic as the in-app agent — an external agent gets exactly the permissions the key's owner has, no more.

get_usage

Remaining quota

list_projects

Projects in the workspace

request_upload_url

Signed upload URL

upload_asset

score_creative

Run RoastIQ

check_status

Poll a running job

get_report

Read a completed report

compare_creatives

Compare 2+ scored ads

search_benchmarks

Benchmark stats by slice

list_personas

Browse BuyerLens personas

run_buyerlens

Synthetic buyer interviews

list_workspaces

Workspaces this key can act in

create_project

Create a new project

FAQ

Before you connect it

Does this cost extra?＋

No separate fee. MCP calls draw from the same monthly analysis allowance as your plan (Free or Pro) — call get_usage before a big upload to check what's left.

Is this a different scoring model than the web app?＋

No. Same pipeline, same benchmark pool, same five KPIs — the MCP server is a thin protocol shim over the identical tool dispatch the in-app agent uses.

Is my creative sent anywhere new?＋

No. The asset goes straight to your existing SaliencyLab workspace, authenticated by your API key. Test keys (sl_test_) are refused here — only a live key tied to your workspace can call it.

What if I don't use Claude?＋

The server speaks standard MCP over Streamable HTTP, so any MCP-capable client can connect in principle — Claude is the client we've built and tested this against. The web app at saliencylab.com works the same regardless.

Score your first ad from inside Claude.

Generate a live API key and connect in about ninety seconds.

Get your API key →