MCP Integration
The Model Context Protocol (MCP) is an open standard that allows AI assistants to communicate with external tools. Omniscribe uses MCP to receive real-time status updates from your AI coding sessions, so you always know what each session is doing without reading the terminal output.
How It Works
Omniscribe bundles its own MCP server -- no separate installation or configuration is needed. When you launch a session, Omniscribe generates a .mcp.json configuration file in the project directory so the AI assistant (such as Claude Code) discovers and connects to the MCP server automatically.
Once connected, the AI assistant can call two MCP tools to report its progress back to Omniscribe.
MCP Tools
omniscribe_status
Reports the current state of the session along with a human-readable message describing what the agent is doing.
Available states:
| State | Meaning |
|---|---|
idle | Waiting for input, not actively working |
working | Executing a task |
planning | Analyzing or planning before implementation |
needs_input | Waiting for clarification from the user |
finished | Task completed successfully |
error | Something went wrong |
Status updates appear as colored indicators in the terminal header, giving you an at-a-glance view of every session's state.
omniscribe_tasks
Reports a structured task list with individual progress tracking. Each task includes an ID, a subject line, and a status (pending, in_progress, or completed).
When an AI assistant reports tasks, a badge appears on the terminal header showing overall progress. Click the badge to expand the full task list and see exactly what the assistant is working on.
Not all AI assistants support MCP tools. Currently, Claude Code has built-in MCP support and will use these tools automatically when connected to Omniscribe.
Configuration
Omniscribe handles MCP configuration automatically per session. You can review discovered MCP server configurations in Settings > MCP Servers.
The generated .mcp.json file is placed in your project root when a session starts. If your project already has an .mcp.json with other MCP servers configured, Omniscribe merges its configuration alongside them.
The MCP server runs locally on your machine. No data is sent to external services -- all communication happens between the AI assistant's terminal process and the Omniscribe app.