An MCP server with 40+ tools, AI call scoring APIs, sales roleplay infrastructure, and signed webhooks. Connect Claude, ChatGPT, or any agent framework in minutes — and ship voice AI, coaching scorecards, and revenue automation without building any of it yourself.
Dialfyne runs voice AI receptionists, sales roleplay training, AI call scoring, and revenue operations in production. The builder platform exposes those same pieces as tools your agents can call.
Score any sales call transcript with an AI coaching engine: 1–10 scores for discovery, objection handling, closing, and rapport, plus specific strengths, missed opportunities, and a next drill. Works on transcripts from any dialer, recorder, or chat.
The killer feature for coaching products: automatically compare every real call against the rep’s roleplay practice history. See improved / consistent / regressed verdicts per skill, and whether reps apply what coaching flagged in practice.
Create roleplay rooms programmatically — persona, scenario, rubric, difficulty, voice, optional realtime video avatar — attach skills and knowledge sources, and read back transcripts and scorecards. Build training products without building a voice stack.
Check availability and book appointments on the same Google Calendar the Dialfyne voice agent uses. Push leads into the workspace queue from your own forms, agents, or enrichment pipelines.
Your agent proposes SMS, calls, emails, or tasks with full context; a human approves in the Command Center before anything executes. Autonomous-agent-safe by design — no accidental outbound, no compliance surprises.
Signed webhooks for call.completed, call.scored, roleplay.session.scored, and lead.created — with retries and delivery logs. React in Slack, your CRM, or your own agent loop the moment something happens.
Dialfyne is a remote MCP (Model Context Protocol) server over streamable HTTP. Claude Code, Claude Desktop, ChatGPT apps, OpenAI Agents, LangChain, and custom frameworks all connect with one URL and a workspace-scoped API key — no SDK, no OAuth dance, no infrastructure.
# Connect from Claude Code — one line
claude mcp add --transport http dialfyne \
"https://voice.dialfyne.com/mcp?clientId=<workspace>&apiKey=<key>"
# Or any MCP client with Bearer auth
{ "mcpServers": { "dialfyne": {
"url": "https://voice.dialfyne.com/mcp",
"headers": { "Authorization": "Bearer <your-api-key>" }
} } }Your exact connection URL — with credentials pre-filled — lives on the Developers page of your Dialfyne dashboard, alongside the full tool catalog and webhook reference.
Point your dialer or call recorder at one endpoint and every connected cold call gets a coaching scorecard automatically — rated on the same rubrics and skills the rep practices in roleplay.
# Auto-score every connected cold call
curl -X POST https://voice.dialfyne.com/webhooks/live-call \
-H "Authorization: Bearer <your-api-key>" \
-d '{
"transcript": "<full call transcript>",
"repEmail": "[email protected]",
"source": "dialer"
}'
# Response: a coaching scorecard + practice comparison
{
"scorecard": { "overall_score": 7, "discovery_score": 8, ... },
"practiceComparison": {
"dimensions": {
"overall_score": { "practiceAverage": 5.5, "live": 7,
"delta": 1.5, "verdict": "improved" }
},
"practiceFocusAreas": ["Ask for the meeting directly"]
}
}// Dialfyne calls YOUR workflow when things happen
POST https://your-agent.example.com/hooks
X-Dialfyne-Event: call.scored
X-Dialfyne-Signature: sha256=<hmac of timestamp.body>
{
"event": "call.scored",
"data": { "repEmail": "jamie@...", "scorecard": { ... } }
}
// Events: call.completed · call.scored ·
// roleplay.session.scored · lead.createdMost platforms only let you call them. Dialfyne also calls you: register a webhook from your agent (it’s a tool call) and your workflow gets signed, retried, logged event deliveries the moment a call ends, a scorecard lands, a practice session is scored, or a lead arrives.
Every payload carries an HMAC-SHA256 signature with a per-subscription secret, deliveries retry with backoff, and a delivery log tool makes debugging integrations trivial.
| Capability | Example tools |
|---|---|
| Coaching & scoring | score_transcript · score_live_call · get_rep_progress · list_live_call_scorecards |
| Roleplay training | create_room · create_skill · create_knowledge_source · list_sessions · get_session |
| Voice AI | list_calls · get_call · get_voice_analytics · get_ai_config |
| Calendar & leads | check_calendar_availability · book_calendar_appointment · create_lead |
| Revenue operations | get_revenue_command_center · upsert_revenue_object · create_revenue_signal · propose_revenue_action |
| Webhooks | register_webhook · list_webhooks · list_webhook_deliveries |
40+ tools total. Your agent can discover the complete list and exact input schemas by listing tools on the MCP server.
Giving an autonomous agent the keys to outbound calling and SMS is how compliance incidents happen. Dialfyne’s permission model is designed for exactly this: agents can read, record, score, and propose freely — but outbound contact always lands in an approval-first queue where a human signs off before anything sends. Combined with workspace-scoped keys, per-endpoint rate limits, signed webhooks, and SMS consent tracking baked into the data model, you get agent autonomy without agent liability.
The Dialfyne builder platform lets developers and AI agent builders embed Dialfyne’s production infrastructure — voice AI calling, AI call scoring, sales roleplay training, calendar booking, and revenue operations — into their own agents and workflows. It is exposed as a remote MCP (Model Context Protocol) server with 40+ tools, a live call scoring API, and signed outbound webhooks. If you are building an AI SDR, a sales coaching product, a CRM copilot, or an internal automation, you can use Dialfyne’s pieces instead of building voice, scoring, and training infrastructure from scratch.
MCP (Model Context Protocol) is the open standard for giving AI models access to external tools and data. Dialfyne hosts a remote MCP server over streamable HTTP, so anything that speaks MCP can connect: Claude (Claude Code, Claude Desktop, the Claude API), ChatGPT apps, OpenAI Agents, LangChain, and custom agent frameworks. Authentication is a single workspace-scoped API key — connect in under a minute with one URL, no SDK installation required.
Yes. The score_transcript tool and the live-call scoring endpoint accept transcripts from any source — your dialer, Zoom, Gong exports, a meeting recorder, even chat logs. Dialfyne’s coaching engine returns a structured scorecard with 1–10 scores for discovery, objection handling, closing, and rapport, plus specific strengths, missed opportunities, and a recommended next drill. You can score against your own rubric, a roleplay room’s rubric, or the skills your team actually trains on.
When you score a real call and include the rep’s email, Dialfyne matches it against that rep’s roleplay practice history and returns a per-dimension comparison: practice average vs live score, the delta, and an improved / consistent / regressed verdict. It also surfaces the focus areas coaching repeatedly flagged in practice, so you can see whether the rep actually applied what they trained on. This turns "did the training work?" from a gut feeling into a measurable answer — automatically, on every connected call.
Reads and recordings are open: your agent can score calls, create leads, build roleplay scenarios, and read analytics directly. Booking on the connected Google Calendar is a direct action. Outbound contact — SMS, calls, emails — always goes through Dialfyne’s approval-first action sequencer: your agent proposes the action with full context, and a human approves it in the Command Center before anything sends. This makes Dialfyne safe to wire into autonomous agents without risking compliance incidents.
Your workspace can register webhook subscriptions (via the register_webhook tool) for call.completed, call.scored, roleplay.session.scored, and lead.created events. Dialfyne POSTs JSON to your HTTPS endpoint with an HMAC-SHA256 signature, retries failed deliveries with backoff, and logs every attempt so you can debug deliveries with the list_webhook_deliveries tool. This lets your workflows react in real time — post scorecards to Slack, update your CRM when a call books, or trigger follow-up sequences when a lead arrives.
Yes. Every API key is scoped to exactly one workspace. Tools, webhooks, scorecards, and data access are all filtered server-side by workspace — an agent holding your key can never read or write another customer’s data. Rate limits apply per endpoint, webhook payloads are signed per subscription with a secret shown only once, and all credentials can be rotated.
Builder access is included with Dialfyne workspace plans — there is no separate API tier or per-call API pricing today. If you are building something for your own team, any plan works. If you are an agency or product team that wants to embed Dialfyne for your customers, book a demo and we will set up partner terms.
Every Dialfyne workspace includes a Developers page in the dashboard with your workspace ID, API key, ready-made MCP connection URL, live-call scoring endpoint, and copy-paste connection snippets for Claude Code and any MCP client. Sign in at app.dialfyne.com, open Developers in the sidebar, and you can have an agent calling Dialfyne tools within five minutes.
Your API key, MCP connection URL, full tool docs, and webhook reference are waiting on the Developers page of your dashboard. Building for clients or embedding Dialfyne in a product? Let’s talk partner terms.