Neuratel MCP Server
Control your entire Neuratel workspace through natural language — from any MCP-compatible AI assistant.
Code
Add to Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):
Code
Then ask Claude: "Create a support agent and call +14155550100" — done.
PyPI · GitHub · Full setup guide
What you can do
- "Create a customer support agent that handles returns and refunds" → agent is created with your instructions, ready to take calls
- "Call +14155550100 using the sales agent" → call is placed immediately
- "What happened on the last call? Show me the transcript" → full conversation returned, turn by turn
- "Start the Q2 outreach campaign" → campaign begins dialing
- "Which calls are active right now?" → live call list with duration and agent info
- "Add our FAQ document as a knowledge source for the support agent" → document indexed and attached
46 focused tools
Most voice AI platforms auto-generate one tool per API endpoint — dumping hundreds of operations on the LLM at once. LLM performance degrades significantly above 30–40 tools.
Neuratel ships 46 hand-crafted tools covering exactly what an AI assistant needs. Every tool is built around a real workflow, not an API endpoint. Every destructive or expensive operation carries an explicit cost warning so your AI assistant never places a call or starts a campaign by accident.
| Domain | Tools | What you can do |
|---|---|---|
| Agents | 8 | Create, inspect, update, duplicate, delete agents; list templates; get required variables |
| Calls | 5 | Place calls, read transcripts, monitor live, hang up |
| Campaigns | 6 | Full campaign lifecycle — create through stop |
| Numbers | 3 | List numbers, assign and unassign agents |
| Knowledge Base | 4 | Add content, list sources, attach to agent |
| Billing | 2 | Check balance, view usage |
| Webhooks | 2 | Create and list event subscriptions |
| Conversations | 8 | Unified inbox: list threads, read/send messages, manage variables, chat analytics |
| DNC | 6 | Check numbers, manage entries, configure protection settings |
| Analytics | 1 | Combined voice + chat KPI dashboard |
| Variables | 1 | System template variable catalog |
Where Neuratel wins
Full call transcripts in the MCP. get_call returns the complete conversation — every turn, every word. Ask "What did the caller say?" and get an actual answer. Neither Vapi nor Retell return transcripts from their MCP tools.
Live call supervision. get_active_calls shows every call in progress right now — who's on, which agent, how long it's been running. Real-time visibility from your AI assistant.
Campaign management. Create a campaign, start it, pause it, check progress, stop it — all from natural language. Vapi doesn't have campaigns. Retell's MCP has no campaign tools.
Knowledge base management. Add documents from text or URLs, list what exists, attach sources to specific agents. No other voice AI MCP server has this.
Cost warnings on everything dangerous. Every tool that places a call, starts a campaign, or terminates something includes an explicit warning in the tool description. Your AI assistant reads these and asks before acting.
Supported clients
| Client | Transport | Notes |
|---|---|---|
| Claude Desktop | stdio | Full tool support |
| Claude Code | stdio | Works in the terminal |
| Cursor | stdio | Add to .cursor/mcp.json |
| Windsurf | stdio | Add to Codeium MCP config |
| Custom agents | streamable-http / SSE | --transport http flag |
Requirements
- Python 3.11+
- Neuratel API key (
nk_live_...) from Settings → API Keys uv(recommended) orpip
→ Setup guide · Tools reference
