# CallSphere MCP Server

A public, read-only [Model Context Protocol](https://modelcontextprotocol.io) server giving any AI assistant — Claude, Claude Desktop, Cursor, Cline, Continue, Windsurf, Zed, or ChatGPT via JSON-RPC — direct access to CallSphere's voice-AI knowledge base, industry ROI calculator, case studies for 6 verticals, and product facts.

- **Endpoint**: `https://callsphere.ai/api/mcp`
- **Transport**: Streamable HTTP
- **Manifest**: https://callsphere.ai/.well-known/mcp.json
- **Docs**: https://callsphere.ai/mcp
- **Auth**: None — free, public, read-only
- **Rate limit**: 120 req/min/IP

## Tools

| Tool | Description |
|------|-------------|
| `search_voice_ai_docs` | Full-text search across 500+ articles on voice AI, MCP, Claude, agentic AI, Twilio/WebRTC, and how-to guides. |
| `read_doc` | Fetch the full markdown body of a single CallSphere blog post by slug. |
| `calculate_voice_ai_roi` | Industry-aware ROI math: monthly savings, payback period, recommended plan tier. Returns assumptions so the AI can show its work. |
| `get_industry_case_study` | Narrative case study for one of 6 verticals (dental, healthcare, HVAC, real estate, salon, law). |
| `get_callsphere_info` | Canonical lookup: pricing tiers, integrations, compliance posture, supported languages, founder contact. |

## Install

### Claude Code

```bash
claude mcp add --transport http callsphere https://callsphere.ai/api/mcp
```

### Claude Desktop

Edit `claude_desktop_config.json` (Settings → Developer → Edit Config) and merge:

```json
{
  "mcpServers": {
    "callsphere": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://callsphere.ai/api/mcp"]
    }
  }
}
```

The `mcp-remote` bridge is used for compatibility with Claude Desktop builds that don't yet speak Streamable HTTP directly. Newer builds can use the simpler:

```json
{
  "mcpServers": {
    "callsphere": {
      "url": "https://callsphere.ai/api/mcp"
    }
  }
}
```

### Cursor

Settings → MCP → **Add new MCP server**, then paste:

```json
{
  "mcpServers": {
    "callsphere": {
      "url": "https://callsphere.ai/api/mcp",
      "transport": "http"
    }
  }
}
```

### Cline / Continue / Windsurf

Add to your MCP servers config:

```json
"callsphere": {
  "url": "https://callsphere.ai/api/mcp",
  "type": "http"
}
```

### ChatGPT / any HTTP client (no MCP support)

The same endpoint accepts plain JSON-RPC 2.0 POST requests:

```bash
curl -sS https://callsphere.ai/api/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"calculate_voice_ai_roi",
                 "arguments":{"industry":"dental","monthly_calls":1200}}}'
```

## Try these prompts

- *"Search CallSphere for articles about building MCP servers with Twilio."*
- *"Calculate the voice-AI ROI for a 4-provider dental practice doing 1,500 calls a month — show your assumptions."*
- *"Show me a case study for HVAC after-hours dispatch."*
- *"Compare CallSphere pricing tiers and tell me which fits 8,000 calls a month."*
- *"What languages does CallSphere support, and is it HIPAA-ready?"*

## Discovery

This server publishes a machine-readable manifest at `/.well-known/mcp.json` so MCP registries (mcp.so, Glama.ai, etc.) can auto-discover and index it. See [`manifest.json`](./manifest.json) in this repo for an offline copy.

## What's NOT in this server (yet)

Phase 2 write tools — outbound demo calls, newsletter signup, demo booking — are intentionally paused. They would incur per-call Twilio/SES costs on what is currently free public infrastructure. They'll arrive when traffic justifies the cost, with explicit per-tool user consent and verification.

## License

MIT for the install snippets and examples in this repo. The MCP server itself is hosted by CallSphere LLC; tool outputs are produced from CallSphere's own data and product configuration.

## Maintained by

[CallSphere LLC](https://callsphere.ai) — Sagar Shankaran ([sagar@callsphere.ai](mailto:sagar@callsphere.ai))