Runs Cloudflare's Clef-Flash decision model locally via llama.cpp so coding agents get probability distributions over their options through one tool.
Setup
Choose your client, copy the config, then fill in any required environment variables.
claude mcp add clefmcp -- npx -y clef-mcp~/.codex/config.toml
[mcp_servers.clefmcp]
command = "npx"
args = ["-y", "clef-mcp"]
# codex mcp add clefmcp -- npx -y clef-mcp~/.cursor/mcp.json
{
"mcpServers": {
"clefmcp": {
"command": "npx",
"args": [
"-y",
"clef-mcp"
]
}
}
}.vscode/mcp.json
{
"servers": {
"clefmcp": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"clef-mcp"
]
}
}
}~/.gemini/settings.json
{
"mcpServers": {
"clefmcp": {
"command": "npx",
"args": [
"-y",
"clef-mcp"
]
}
}
}claude_desktop_config.json
{
"mcpServers": {
"clefmcp": {
"command": "npx",
"args": [
"-y",
"clef-mcp"
]
}
}
}.trae/mcp.json
{
"mcpServers": {
"clefmcp": {
"command": "npx",
"args": [
"-y",
"clef-mcp"
]
}
}
}