mcp-signal
Local Signal MCP server: read via Signal Desktop, send via signal-cli
README
mcp-signal
A local Model Context Protocol (MCP) server that reads Signal Desktop history from the local encrypted database via
signal-exportand sends outbound messages viasignal-cli.
mcp-signal focuses on the core workflow for personal Signal automation — list chats, read messages, search messages, inspect groups, and send messages to direct or group chats. Everything runs locally; stdio transport with no network listener.
Heads up — mixed backend. Read/search comes from the local Signal Desktop database. Sending uses
signal-cli, which must be installed and linked to a Signal account separately. Ifsignal-cliis unavailable, read/search still works but send tools do not.
Features
- List direct and group chats from Signal Desktop
- Read recent messages from a chat
- Search messages within one chat or across all chats
- List group chats with
signal-cligroup IDs for outbound use - Send a message to:
- a direct recipient by phone number
- a group by group ID
- a chat by exact chat name (with ambiguity checks)
- Runs entirely on your machine; stdio transport with no network listener
Setup
Prerequisites
- Python 3.13+
- uv
- Signal Desktop with an existing local message database
signal-cliinstalled and linked if you want outbound sends
Installation
-
Clone this repository
git clone https://github.com/Sealjay/mcp-signal.git cd mcp-signal -
Install dependencies
uv sync -
Install
signal-cli(optional — only needed for outbound sends)On macOS, the simplest route is Homebrew:
brew install signal-cli
Configure outbound sends
The server auto-loads a local .env.local file from the repo root if present. This file is gitignored and is the recommended place for machine-local config.
cat > .env.local <<'EOF'
SIGNAL_ACCOUNT="+441234567890"
EOF
Optional environment variables:
| Variable | Purpose |
|---|---|
SIGNAL_CLI_PATH |
Override the signal-cli binary path |
SIGNAL_DATA_DIR |
Override the Signal Desktop data directory |
SIGNAL_DB_PASSWORD |
Password for encrypted desktop DBs if needed |
SIGNAL_DB_KEY |
Raw key for encrypted desktop DBs if needed |
Environment variables set in the shell take precedence over .env.local.
Link signal-cli (first run only)
mcp-signal does not manage linking itself. Link the local signal-cli device first:
signal-cli link -n "signal-mcp"
Scan the QR code in the Signal mobile app (Settings → Linked Devices → Link New Device).
Do not pass -a / --account to link on current signal-cli versions — linking a new secondary device does not take a phone number there.
After the QR is accepted, confirm the linked account is visible:
signal-cli listAccounts
That account should match the SIGNAL_ACCOUNT value in .env.local.
signal-cli stores its linked-account state under its own local data directory (typically ~/.local/share/signal-cli/data on macOS/Linux). That state lives outside this repository and is not committed by mcp-signal.
Verify everything is connected:
uv run signal-mcp smoke
MCP client configuration
All clients launch the server the same way over stdio. On macOS, you may need the absolute path to uv — see macOS: uv PATH below.
Claude Code
The quickest route is the CLI:
claude mcp add --transport stdio signal --scope user -- uv run --directory /absolute/path/to/mcp-signal signal-mcp serve
Alternatively, add to .mcp.json at your project root (or ~/.claude.json for a user-scoped server):
{
"mcpServers": {
"signal": {
"type": "stdio",
"command": "uv",
"args": ["run", "--directory", "/absolute/path/to/mcp-signal", "signal-mcp", "serve"]
}
}
}
If you edit the file directly, restart the Claude Code session to pick it up.
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):
{
"mcpServers": {
"signal": {
"command": "uv",
"args": ["run", "--directory", "/absolute/path/to/mcp-signal", "signal-mcp", "serve"]
}
}
}
Restart Claude Desktop. You should see signal listed as an available integration.
Cursor
Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"signal": {
"command": "uv",
"args": ["run", "--directory", "/absolute/path/to/mcp-signal", "signal-mcp", "serve"]
}
}
}
Restart Cursor.
macOS: uv PATH
GUI apps (Claude Desktop, Cursor) don't always inherit the PATH from your interactive terminal, so uv may fail with spawn uv ENOENT. Fix by using the absolute path to uv in command:
- Homebrew —
/opt/homebrew/bin/uv(Apple Silicon) or/usr/local/bin/uv(Intel) - Manual install — run
which uvin your terminal to find it
Example:
{
"mcpServers": {
"signal": {
"command": "/opt/homebrew/bin/uv",
"args": ["run", "--directory", "/absolute/path/to/mcp-signal", "signal-mcp", "serve"]
}
}
}
Architecture
| Component | Description |
|---|---|
| MCP server | Python/FastMCP, stdio transport |
| Read path | signal-export reading the local Signal Desktop database |
| Send path | signal-cli JSON-RPC launched on demand |
| State | No separate cache; reads directly from Signal Desktop data |
Data flow
- The MCP client launches
signal-mcp serveover stdio. - Read/search tools call
signal-exportagainst the local Signal Desktop database. - Group listing and outbound sends call
signal-cli -a ACCOUNT jsonRpc. - Results are returned as structured JSON.
Project structure
mcp-signal/
src/mcp_signal/
config.py
main.py
reader.py
server.py
signal_cli.py
tests/
CLAUDE.md
LICENSE
README.md
SECURITY.md
Tools
| Tool | Purpose |
|---|---|
list_chats |
List direct and group chats from Signal Desktop |
read_messages |
Read messages from a specific chat |
search_messages |
Search messages within one chat or across all chats |
list_groups |
List groups from signal-cli, including group IDs |
send_message |
Send a text message to a direct recipient or group |
get_status |
Show desktop DB / signal-cli / account readiness |
Privacy and security
- No cloud relay. No network listener. All data stays on your machine.
- Read/search uses your local Signal Desktop data only.
- Send operations require a locally configured
signal-cliaccount. .env.localis intended for local secrets such asSIGNAL_ACCOUNTand is not committed.signal-clilinked-device state is stored in its own local app data directory, outside this repo, and is not committed.
See SECURITY.md for how to report vulnerabilities.
Limitations
- Prompt-injection risk: as with many MCP servers, this one is subject to the lethal trifecta. Malicious incoming messages could attempt to instruct an agent to exfiltrate other messages. Treat the tool surface accordingly and review outbound actions before approving them.
- Mixed backend: chat history comes from Signal Desktop, while outbound sends come from
signal-cli. - No attachments: text-only send.
- No real-time notifications: polling/read only.
- Single account per MCP instance.
- Group sends need
signal-cli: local DB reads alone do not provide enough information to send to groups safely.
Development
uv sync
uv run signal-mcp smoke
uv run pytest
uv run ruff check .
Troubleshooting
signal-clinot found — confirmsignal-cliis onPATHor setSIGNAL_CLI_PATHin.env.local. On macOS,brew install signal-cliis the simplest route.- Read/search works but sends fail —
signal-cliis not linked orSIGNAL_ACCOUNTis not set. Runsignal-cli listAccountsto verify, then check.env.local. signal-cli linkhangs or fails — do not pass-a/--accounttolinkon current versions. Runsignal-cli link -n "signal-mcp"and scan the QR from your phone.- MCP client can't launch the server —
argsmust contain an absolute path to the repo, not relative. Ifuvitself fails withspawn uv ENOENT, see macOS:uvPATH. - No messages returned — confirm Signal Desktop is installed and has message history. The read path queries the local Signal Desktop database directly.
Contributing
Contributions welcome via pull request. Please:
- Run
uv run ruff check .before pushing. - Ensure
uv run pytestpasses.
See CLAUDE.md for the full development workflow.
Licence
MIT Licence — see LICENSE.
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。