Meridian
Human-in-the-loop prompting for coding agents — approve, choose, fill a form, or notify.
Connect your client
Claude Code adds it straight from the CLI. In Claude Desktop or claude.ai, open Settings → Connectors → Add custom connector and paste the endpoint URL.
claude mcp add --transport http meridian https://mcp.fastverk.com/meridian/mcp{
"mcpServers": {
"meridian": {
"url": "https://mcp.fastverk.com/meridian/mcp"
}
}
}{
"servers": {
"meridian": {
"type": "http",
"url": "https://mcp.fastverk.com/meridian/mcp"
}
}
}Codex CLI speaks stdio, so it bridges to the hosted endpoint with mcp-remote.
[mcp_servers.meridian]
command = "npx"
args = ["-y", "mcp-remote", "https://mcp.fastverk.com/meridian/mcp"]Add it under Windsurf → Settings → MCP, or edit the config directly.
{
"mcpServers": {
"meridian": {
"serverUrl": "https://mcp.fastverk.com/meridian/mcp"
}
}
}Add it to Zed's settings.json — it bridges to the hosted endpoint with mcp-remote.
{
"context_servers": {
"meridian": {
"source": "custom",
"command": "npx",
"args": ["-y", "mcp-remote", "https://mcp.fastverk.com/meridian/mcp"]
}
}
}First connect triggers sign-in if the server requires it.
About
Meridian lets a coding agent pause and ask a human. Four tools: request_approval (approve/reject a proposed diff), ask_choice (pick one of several options), request_input (fill a structured form), and notify (fire-and-forget status). Each prompt is written as a HumanPrompt CR and rendered as a meridian component in the console; the human's answer flows back to the agent — the blocking tools wait on it over NATS, or hand back a durable pending handle to resume from.