@disciplog/mcp

The DiscipLog Model Context Protocol (MCP) Server

This MCP server provides a standardized bridge between your AI coding agents (like Claude Code, Cursor, Antigravity) and your DiscipLog productivity coach.

Why does this exist?

DiscipLog isn't just a time tracker—it's an AI productivity coach. To coach you effectively on how you interact with AI, DiscipLog needs to understand the context of your prompting. Instead of intercepting your private code, this MCP server allows AI agents to securely report metadata about your coding sessions.

The agent uses this server to log:

• Prompt Characteristics: Was your prompt clear? Did you provide context, constraints, or examples?
• AI Turn Summaries: A brief narrative of what the AI decided to do (e.g., "Refactored the auth module and added tests").
• Agent Self-Critique: A highly valuable signal where the AI rates your prompt and suggests how you could have been more explicit.
• Timing & Latency: How long you spent thinking vs. typing vs. waiting for the AI to respond.

This metadata feeds directly into your DiscipLog dashboard, allowing your AI Coach to spot over-reliance on AI, repeated prompting mistakes, or strong iteration cycles.

Installation

One command (recommended)

npx @disciplog/mcp install

Detects which AI clients are installed on your machine (Claude Code, Codex, Cursor, Claude Desktop, Antigravity) and lets you pick where to register DiscipLog — with an "install everywhere" shortcut. After it finishes you'll see a per-client restart hint (because every MCP client reads its config at app launch).

Flags:

npx @disciplog/mcp install --all --yes        # install everywhere, no prompts
npx @disciplog/mcp install --client claude-code --client cursor
npx @disciplog/mcp install --api-key dlog_live_…  # skip the API-key prompt

Removing DiscipLog mirrors install:

npx @disciplog/mcp uninstall                  # interactive
npx @disciplog/mcp uninstall --all --yes      # remove everywhere
npx @disciplog/mcp uninstall --all --purge --yes  # also wipe ~/.disciplog/

Check status at any time:

npx @disciplog/mcp doctor

The uninstaller only ever removes entries the installer placed (every config entry it writes carries _disciplogManaged: true, or for Codex's TOML a # managed-by: disciplog comment). User-authored MCP servers and hooks in the same config files are left untouched.

Manual MCP config

If you'd rather edit your client's MCP config by hand, the entry is:

{
  "mcpServers": {
    "disciplog": {
      "command": "npx",
      "args": ["-y", "@disciplog/mcp"],
      "env": {
        "DISCIPLOG_API_KEY": "dlog_live_...",
        "DISCIPLOG_API_URL": "https://disciplog.com"
      }
    }
  }
}

For Codex (TOML):

[mcp_servers.disciplog]
command = "npx"
args = ["-y", "@disciplog/mcp"]
env = { DISCIPLOG_API_KEY = "dlog_live_...", DISCIPLOG_API_URL = "https://disciplog.com" }

Via the VS Code extension

If you use the DiscipLog VS Code Extension, DiscipLog: Connect will paste your API key and register DiscipLog with detected AI agents automatically.

Available Tools

The MCP server exposes the following tools to the LLM agent:

• start_ai_turn: Called immediately when processing a new prompt to track precise response-latency timing.
• log_ai_turn: Called at the end of the response to submit the prompt characteristics, summary, and self-critique.
• start_ai_blocked_on_user / finish_ai_blocked_on_user: Called when the AI pauses to ask for your permission (e.g., before running a destructive terminal command), ensuring this wait time isn't counted against your productive hours.
• start_ai_waiting_on_user / finish_ai_waiting_on_user: Called when the AI pauses for a non-permission user decision, clarification, credential, or account-setup step, keeping normal collaboration waits separate from permission gates.

Privacy First

This server never transmits your actual code or verbatim prompts to DiscipLog. It only transmits the structural metadata and generalized summaries generated by the AI agent.