Skip to main content
RoastIQBuyerLensHugoPricingBlogAbout
Client accessDiscuss your campaign

SaliencyLab MCP & CLI, for the AI client you already use.

Use your preferred AI client to explore creative evidence, run diagnostics and simulate buyer questions. MCP and CLI are optional paths into the same workspace.

MCP and CLI use your workspace permissions and analysis allowancesOPTIONAL AGENT ACCESS
1

Copy the connector URL

You'll paste this into Claude in the next step.

https://mcp.saliencylab.com
2

Go to Claude → Connectors

In Claude Desktop or claude.ai, go to Settings → Connectors → Add custom connector. Name it SaliencyLab and paste the URL.

3

Connect and ask

Paste your sl_live_ key when prompted, then ask Claude to score an ad.

Start scoring
If you're using Claude Code or another terminal-based agent, it's better to use the CLI toggle ↗OpenAPI spec ↗

How does scoring work?

Four things an agent asks before it ships. Every call runs through the same auth, quota, and workspace-tenancy logic as the in-app agent.

Claude
The client wants to know if the 15s TikTok cut is safe to launch before tomorrow's media buy locks. I'll score it against the DTC beauty benchmark first.
RoastIQ
score_creative(asset_id: "ttk_0231", category: "dtc-beauty", platform: "tiktok")
TikTok9:16DTC beauty
✓ Result
Glow Drops · TikTok 15s cutdownSHARPEN
54Skip
76Notice
80Brand
70Sell
66Build
Percentile
74th+12 vs avg
Benchmark
n=214v4 pool
Benchmark v4 · high confidence

14 tools. One dispatch layer.

An external agent gets exactly the permissions its key's owner has, no more. Click a card to see what it returns; on a phone, swipe to see the rest.

get_usage

Remaining quota

Click to preview ↻
68% of monthly quota used
list_projects

Projects in the workspace

Click to preview ↻
3 projects in this workspace
request_upload_url

Signed upload URL

Click to preview ↻
One signed URL, 15 minute expiry
upload_asset

Register a scoreable asset

Click to preview ↻
Asset registered, ready to score
score_creative

Run RoastIQ

Click to preview ↻
Composite scored in about 70 seconds
check_status

Poll a running job

Click to preview ↻
Queued, then processing, then done
get_report

Read a completed report

Click to preview ↻
Full KPI breakdown, one call
compare_creatives

Compare 2+ scored ads

Click to preview ↻
Cut B wins on 2 of 3 KPIs
search_benchmarks

Benchmark stats by slice

Click to preview ↻
Where this ad sits vs. the pool
list_personas

Browse BuyerLens personas

Click to preview ↻
6–12 personas per study
run_buyerlens

Synthetic buyer interviews

Click to preview ↻
Six simulated buyer interviews
list_workspaces

Workspaces this key can act in

Click to preview ↻
Every workspace this key can touch
create_project

Create a new project

Click to preview ↻
Starts empty, ready for the first roast
get_panel_results

BuyerLens recap for agents

Click to preview ↻
Objections by prevalence, every quote labelled synthetic

Questions before you connect.

Whatever your team already has open. Claude Desktop and Claude Code are the two we've built and tested end-to-end. Any MCP-compatible client — ChatGPT, Cursor, and others — can connect in principle; we've only tested Claude ourselves so far.

Get your API key →