Killbottleneck

What killbottleneck MCP server does

The killbottleneck MCP server gives AI agents a way to work with a self-hosted killBottleneck workspace. The application is built around visual goal maps: users can create nodes and edges, add statuses and notes, maintain multiple maps, comment on goals, and share maps with read or edit access. Public maps and exports to image or PDF are also available in the web application.

The product is intended for joint work by people and agents. Its broader workspace includes projects, processes, tasks, clients, time records, messages, and documents. A mobile-oriented view supports reviewing and completing today’s tasks, adding a task, and reading messages.

How it works

killBottleneck runs as an application on your own server. The documented quick start uses Docker Compose and exposes the application on port 8090 by default. The first registered user becomes the administrator; later users can register or receive invitations from the administration area.

The MCP connection is described in the README as an AI-assistant feature, with Claude Desktop and Claude Code named as supported clients. The supplied material does not specify whether the MCP endpoint uses stdio or HTTP, provide a client configuration example, or enumerate the available MCP tool names. Those details should be checked in the repository’s complete documentation before implementation.

Setup and configuration

Start the application by copying the example environment file and launching the Compose stack:

cp .env.example .env
docker compose up -d

The default installation does not require AI or external services. AI is disabled with KB_AI_PROVIDER=none, and the README states that the default server sends no requests to third-party services. Optional AI modes include OpenAI-compatible APIs, Ollama, a compatible remote API, and a custom endpoint. Separate configuration is available for chat, vision, transcription, and summary generation.

For access beyond a local network, the documentation covers HTTPS and reverse proxies. HTTPS is also needed for browser microphone access except on localhost. The application supports SMTP, Google sign-in and Drive picking, attachments, and other optional features when their settings are supplied.

Tools and capabilities

The killbottleneck MCP server is intended for agent access to the same goal and process information that people manage in the application. Grounded capabilities of the underlying workspace include:

• Create and edit structured goal maps with nodes, edges, statuses, and notes.
• Work with projects, tasks, comments, and shared maps.
• Review task timing through timers and time records.
• Use project-to-client relationships for time reporting.
• Combine human work with optional AI planning and map assistance.

The README says AI proposals are presented as cards for confirmation rather than being written without user approval. That behavior describes the built-in assistant; the supplied excerpt does not state whether every MCP mutation follows the same confirmation flow.

Limitations and notes

The project is described as a public beta, including its self-hosted deployment. Installation, reverse proxies, SMTP, and upgrades are still areas being tested. The provided material does not give the MCP server’s complete API surface, authentication model, transport details, or permissions behavior, so those points should not be assumed from the application’s web features alone.

Optional AI changes the data boundary: map data is sent to the endpoint configured by the operator. With AI disabled, the README says data does not leave the server. The software is open in spirit but uses a fair-code license that prohibits reselling it as a hosted service; the excerpt does not provide a short SPDX license name.