@kanopi/callrail-mcp
A Model Context Protocol (MCP) server for the CallRail API v3 that exposes call tracking, text messaging, form submissions, and more as MCP tools for reading and writing to your CallRail account.
README
@kanopi/callrail-mcp
A Model Context Protocol (MCP) server for the CallRail API v3. It exposes CallRail's call tracking, text messaging, form submissions, trackers, companies, tags and more as MCP tools so an LLM client (Claude Desktop, Claude Code, etc.) can read from and write to your CallRail account.
- Transport: stdio (local)
- Auth: CallRail API v3 key via environment variable
- Coverage: 34 tools across 11 resources (full read + write)
Prerequisites
- Node.js 18+ (uses the global
fetch) - A CallRail API v3 key — in CallRail go to Settings → Integrations → API Keys → Create New API v3 Key. Keys are user-scoped; the server can only see what that user can.
Usage
No install needed — run the published package directly with npx:
npx -y @kanopi/callrail-mcp
npx fetches @kanopi/callrail-mcp and runs its callrail-mcp binary over stdio. This
is what you point your MCP client at (see Connecting to a client).
Build from source (for development)
git clone <repo-url> && cd call-rail-mcp
npm install
npm run build
This compiles TypeScript to dist/. The entrypoint is dist/index.js.
Configuration
The server reads credentials from environment variables:
| Variable | Required | Description |
|---|---|---|
CALLRAIL_API_KEY |
✅ | Your CallRail API v3 key. |
CALLRAIL_ACCOUNT_ID |
optional | Default account id. When set, tools may omit account_id. Find it with the list_accounts tool. |
CALLRAIL_BASE_URL |
optional | Override the API base URL (default https://api.callrail.com/v3). Mainly for testing. |
Copy .env.example to .env for local reference (the server reads from the process
environment; your MCP client passes these in — see below).
Connecting to a client
Claude Code
claude mcp add callrail \
--env CALLRAIL_API_KEY=your_api_key_here \
--env CALLRAIL_ACCOUNT_ID=your_account_id \
-- npx -y @kanopi/callrail-mcp
Claude Desktop / other clients (JSON config)
Add to the mcpServers block of the client's config
(e.g. claude_desktop_config.json):
{
"mcpServers": {
"callrail": {
"command": "npx",
"args": ["-y", "@kanopi/callrail-mcp"],
"env": {
"CALLRAIL_API_KEY": "your_api_key_here",
"CALLRAIL_ACCOUNT_ID": "your_account_id"
}
}
}
}
To run a local checkout instead of the published package, point at the built entrypoint —
"command": "node", "args": ["/absolute/path/to/call-rail-mcp/dist/index.js"] — or the
TypeScript source via tsx: "command": "npx", "args": ["tsx", "/abs/path/src/index.ts"].
Tools
Write tools are prefixed with WRITE: in their descriptions and carry MCP annotations
(readOnlyHint/destructiveHint) so clients can prompt for confirmation. Most tools accept
an optional account_id that falls back to CALLRAIL_ACCOUNT_ID. List tools share common
params: page, per_page (max 250), sort, order, search, date_range/start_date/
end_date, and fields for optional extra fields.
| Resource | Tools |
|---|---|
| Accounts | list_accounts, get_account |
| Companies | list_companies, get_company, create_company ✏️, update_company ✏️, delete_company 🗑️ |
| Calls | list_calls, get_call, create_outbound_call ✏️, update_call ✏️, get_calls_summary, get_calls_timeseries |
| Trackers | list_trackers, get_tracker, create_tracker ✏️, update_tracker ✏️, delete_tracker 🗑️ |
| Text messages | list_conversations, get_conversation, send_text_message ✏️ |
| Form submissions | list_form_submissions, get_form_submission, update_form_submission ✏️ |
| Users | list_users, get_user |
| Tags | list_tags, create_tag ✏️, update_tag ✏️, delete_tag 🗑️ |
| Integrations | list_integrations (per company) |
| Notifications | list_notifications |
| Outbound caller IDs | list_outbound_caller_ids, create_outbound_caller_id ✏️ |
✏️ = write (create/update) · 🗑️ = destructive (delete)
Notes & caveats
- Premium data: transcription, sentiment, call summaries and highlights require a
CallRail Premium Conversation Intelligence subscription. Request them via the
fieldsargument onget_call/list_calls. - SMS compliance: outbound texts must include opt-out instructions; CallRail enforces this.
- Rate limits: ~1,000 requests/hour by default. The server surfaces HTTP 429 (with
Retry-Afterwhen provided) as a tool error rather than crashing. - Data retention: communication records are retained ~25 months; older date ranges error.
- Pagination is manual: pass
pageto walk results; the server does not auto-page.
Development
npm run dev # run from source with tsx
npm run build # type-check + emit dist/
npm start # run the built server
Project layout
src/
index.ts entrypoint (stdio transport)
config.ts env loading + validation
client.ts CallRail HTTP client (auth, query, errors, account resolution)
types.ts shared types + CallRailApiError
tools/
index.ts registers all tools
shared.ts common zod params, response/error helpers, annotations
<resource>.ts one file per CallRail resource
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。