Clarity for AI agents · prototype
Your AI agent makes the calls. Clarity finds what works.
A/B testing needs thousands of dials: about two months for a five-rep team, about a week for one AI agent. Connect your agent over MCP or REST and Clarity tells it which script to run, logs every outcome, and learns which behaviors book meetings.
See the difference volume makes
Same two scripts, same statistics. Run it with a human team, then with an agent. The only thing that changes is how fast you get a trustworthy answer.
Who's dialing
True meeting rate (hidden from Clarity)
Press start. Calls split evenly between A and B, and Clarity checks the result every day.
Day
0
2,000 dials/day
plan for ≈ 7 days
How it plugs into your agent
Step 1
assign_variant
Before each call, the agent asks Clarity which script to use. Variants are balanced per agent, so the test stays fair.
Step 2
Your agent calls
Use any voice or SDR agent stack. Clarity never places calls; it decides what to test and learns from the result.
Step 3
log_outcome
Report the outcome and, ideally, the transcript. Clarity measures the behaviors and keeps the numbers, not the words.
6 tools, one MCP server
list_experiments
List the team's running A/B tests and their script variants.
assign_variant
Before a call: get which script variant this agent should use for an A/B test. Balances variants per agent. Returns the full script (opener, branches, objection handlers, voicemail).
log_outcome
After a call: record the outcome. Include the transcript to capture behaviors (talk share, questions, call-flow checks) so Clarity can learn which behaviors book meetings. Only measurements are stored, never the transcript.
get_results
Current results for an A/B test: per-variant dials, connect and meeting rates, verdict at 95% confidence, and dials still needed.
get_playbook
Before a shift or a call: the team's current playbook from real results. Which script to run and why, behaviors that book more (or fewer) meetings per conversation at 95% confidence, the objections you'll most likely hear with the answer to use, and how running tests stand.
analyze_call
Analyze a call transcript without storing anything: talk share, questions, monologues, fillers, objections and call-flow checks.
Connect in a minute
Create a key under Settings → API, then add Clarity to any MCP-compatible agent. Prefer HTTP? Every tool is also POST /api/v1/<tool>.
{
"mcpServers": {
"clarity": {
"type": "http",
"url": "https://claritysdr.com/api/mcp",
"headers": { "Authorization": "Bearer clr_live_…" }
}
}
}Prototype notes
- Agent calls are tagged with their agent name and kept apart from your reps' leaderboards.
- Transcripts are analyzed on the fly and discarded. Only behavior measurements are stored.
- You're responsible for consent and disclosure rules for AI-placed calls where you operate.
Building an AI SDR or voice agent?
We're looking for a few agent teams to shape this API with us.
Talk to the founder