@flarewatch/mcp
stdio→HTTPS bridge for the FlareWatch MCP server.
Lets stdio-only MCP clients (Claude Desktop, Cursor, Continue) reach
the remote endpoint at https://mcp.flarewatch.io/api/mcp without
each client having to implement the Streamable HTTP transport.
Install / use
No install needed — npx will fetch on demand:
npx -y @flarewatch/mcp
Claude Desktop
Edit ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"flarewatch": {
"command": "npx",
"args": ["-y", "@flarewatch/mcp"]
}
}
}
Restart Claude Desktop. The 13 FlareWatch tools (validators, FTSO
providers, scoring methodology, agent transparency, etc.) become
available in every conversation.
Cursor
Cursor supports Streamable HTTP natively — you don't need this bridge.
Configure directly:
{
"url": "https://mcp.flarewatch.io/api/mcp"
}
Continue.dev
{
"experimental": {
"modelContextProtocolServers": [
{
"transport": { "type": "stdio", "command": "npx", "args": ["-y", "@flarewatch/mcp"] }
}
]
}
}
Environment overrides
FLAREWATCH_MCP_URL | https://mcp.flarewatch.io/api/mcp | Endpoint to forward to. Useful for preview deploys or local dev. |
FLAREWATCH_MCP_TIMEOUT_MS | 30000 | Per-request HTTP timeout in ms. |
What this bridge does NOT do
It's intentionally dumb. ~150 LoC. The server is the single source of
truth for:
- Tool definitions
- Resource content
- Rate limiting
- Quarantine / honeypot / classifier walls
- Response signing
The bridge just forwards JSON-RPC frames between stdio and HTTPS. No
caching, no retries, no fallback. If the remote endpoint is down, the
bridge surfaces the upstream error rather than attempting recovery.
Verification
Every tool response carries a verification_signature HMAC. To verify
a citation came from this server, call the verify_response tool with
the cited payload + signature.