EM Copilot

EM Copilot is an AI-powered assistant built into EM Kit. It connects to a local coding agent — Claude Code or Codex CLI — and gives it access to your workspace data through the EM Kit MCP server.

Use it to ask questions about your team, summarize feedback, draft review notes, or surface action items — all from within the app.

Requirements

Before using EM Copilot:

1. Install one of the supported providers on your machine:
   • Claude Code — install from docs.anthropic.com and authenticate in your terminal.
   • Codex CLI — install from openai.com and authenticate in your terminal.
2. Make sure the provider command (claude or codex) is available in your terminal PATH.

Configure EM Copilot

1. Open Settings → EM Copilot.
2. Select your Provider from the dropdown.
3. Optionally set a Custom command path if the provider executable is not on PATH.
4. Optionally pick a Default model and Reasoning effort — these apply to Copilot conversations, generated insights, and automations for the current workspace. Individual automations can still override both.
5. Click Save AI assistant settings.
6. Click Refresh to verify the provider is detected and shows as “Installed”.

The status table below the form shows each provider’s installation status, executable path, and version.

Claude Code connectors

Claude Code is always invoked in fast non-interactive --print mode. EM Kit merges its own MCP server with your user-scope Claude MCP configuration, so any Claude Code connector you registered with claude mcp add -s user … (Slack, Atlassian, Notion, etc.) is automatically available to the Copilot. Run claude mcp list in a terminal to see which connectors Claude currently has access to.

Using the assistant

Once configured, click the Copilot button in the top-right corner of the app to open the assistant drawer.

EM Kit automatically starts your selected provider and connects it to the EM Kit MCP server, giving the assistant access to your workspace data — team members, work log, feedback, daily notes, issues, and more. The assistant can also create and update issues on your behalf.

Agents and skills

Each Copilot conversation runs through an agent — a customizable persona with its own system prompt and an optional set of skills (recipes triggered by phrases like “prep me for a 1:1”). The agent picker sits in the draft bar; switch agents at any time.

Six built-in agents and four built-in skills ship with every workspace, and you can create your own. See Agents and Skills for the full reference.

Example prompts

• “Summarize the feedback I need to share this week.”
• “What are my open issues?”
• “Create an urgent issue to fix the login bug and assign it to me.”
• “Draft a quick status update from this week’s work log entries.”
• “List the team members who have a review coming up.”

Troubleshooting

Provider shows “Not installed”

1. Open a terminal and run which claude or which codex to confirm the CLI is on PATH.
2. If you installed the CLI in a custom location, enter the full path in the Custom command path field.
3. Click Refresh to re-check.

Assistant does not respond

1. Make sure the EM Kit desktop app is running — the MCP server requires it.
2. Run emkit doctor in your terminal to verify the CLI bridge is healthy.
3. Check that you have authenticated the provider CLI at least once in your terminal.

”Claude Code did not respond within 600s”

The provider exceeded its wall-clock budget for a single run. This is usually one of:

1. The CLI is waiting for login — run claude login (or codex login) in your terminal and complete the flow, then try again.
2. The prompt triggered a long tool-calling loop. Try narrowing the question or lowering the reasoning effort.
3. A slow MCP connector is blocking start-up. Run claude mcp list to see which connectors Claude has registered and confirm each one is healthy.

”Claude Code is not authenticated”

Run claude login (or codex login) in your terminal, complete the browser flow, then retry the request from EM Copilot. The Copilot surfaces this as a distinct error — you do not need to restart the app after authenticating.

”Claude Code stopped at the max-turns limit”

The run hit Claude Code’s built-in max-turns limit before producing a final answer. Either simplify the prompt so fewer tool calls are needed, or raise the limit via --max-turns in your Claude Code configuration.