Setup & Connection
Connect your AI coding assistant or agent runtime to the Knowledgator MCP server using either our hosted platform endpoint or by running a local instance.
Connection Modes
| Feature | Hosted Remote Endpoint | Self-Hosted / Local |
|---|---|---|
| Best For | Quickstart, Claude Desktop, Cursor, remote agent runtimes | Local development, custom agent harnesses, private VPCs |
| Prerequisites | None (Zero installation) | Python 3.10+, Docker, or uv |
| Connection Method | Remote MCP (https://mcp.knowledgator.com/mcp) | Streamable HTTP (127.0.0.1:8000/mcp), SSE (/sse), or Stdio |
| Maintenance | Fully managed by Knowledgator | Self-managed dependencies and updates |
| Setup Guide | View Remote Setup → | View Self-Hosted Setup → |
1. Hosted Remote Endpoint (Zero Install)
The fastest way to use the Knowledgator MCP server. Connect directly to Knowledgator's platform infrastructure via MCP without installing Python packages or cloning repositories.
- MCP URL:
https://mcp.knowledgator.com/mcp - Required Header:
KnowledgatorAPI-Key: YOUR_KNOWLEDGATOR_API_KEY
Claude Desktop Configuration
Add the server to your claude_desktop_config.json:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json - Linux:
~/.config/Claude/claude_desktop_config.json
{
"mcpServers": {
"knowledgator": {
"url": "https://mcp.knowledgator.com/mcp",
"headers": {
"KnowledgatorAPI-Key": "YOUR_KNOWLEDGATOR_API_KEY"
}
}
}
}
Cursor Configuration
In Cursor, navigate to Settings > Features > MCP, then click Add New MCP Server:
- Name:
knowledgator - Type:
sse - URL:
https://mcp.knowledgator.com/mcp - Headers: Add
KnowledgatorAPI-Key: YOUR_KNOWLEDGATOR_API_KEY
2. Self-Hosted / Local Server
If you prefer to host the server locally on your machine or deploy it within a private cluster:
Installation Methods
A. Run Directly with uvx (No clone needed)
uvx --from git+https://github.com/knowledgator/knowledgator-mcp.git knowledgator-mcp
B. Install via pip from Git
pip install git+https://github.com/knowledgator/knowledgator-mcp.git
C. Clone & Install with Poetry / Pip
git clone https://github.com/knowledgator/knowledgator-mcp.git
cd knowledgator-mcp
# With Poetry:
poetry install
# Or with pip editable:
pip install -e .
D. Run with Docker
docker build -t knowledgator-mcp .
docker run -e KG_API_KEY="YOUR_KNOWLEDGATOR_API_KEY" -p 8000:8000 knowledgator-mcp --transport streamable-http --host 0.0.0.0 --port 8000
ℹ️ Note:
pip install knowledgator-mcpvia PyPI will be available soon once package publication is complete.
Local Client Setup (Claude Desktop via Stdio)
{
"mcpServers": {
"knowledgator": {
"command": "uvx",
"args": ["--from", "git+https://github.com/knowledgator/knowledgator-mcp.git", "knowledgator-mcp"],
"env": {
"KG_API_KEY": "YOUR_KNOWLEDGATOR_API_KEY"
}
}
}
}
Configuration & Environment Variables
When running locally or self-hosting, configure the server using environment variables or a .env file:
| Variable | Default | Required | Description |
|---|---|---|---|
KG_API_KEY or KG_MCP_API_KEY | (None) | Yes | Your Knowledgator platform API key. |
KG_MCP_BASE_URL | https://api.platform.knowledgator.com | No | Base URL for Knowledgator API requests. Override this for local testing, staging environments, or custom enterprise domains. |
KG_MCP_DEBUG | false | No | Enables verbose debug logging for all requests and tool dispatches. |
KG_MCP_SERVER_NAME | KnowledgatorMCP | No | Identifier name reported to connected MCP clients. |
KG_MCP_REQUEST_TIMEOUT_MS | 60000 | No | Default HTTP timeout in milliseconds (default: 60s). |
Server Transports
The server supports three runtime transports:
# 1. Streamable HTTP (Primary, default - mounts at /mcp)
knowledgator-mcp --transport streamable-http --host 0.0.0.0 --port 8000
# 2. SSE Server (Secondary option - mounts at /sse)
knowledgator-mcp --transport sse --host 0.0.0.0 --port 8000
# 3. Standard I/O (For local CLI clients like Claude Desktop)
knowledgator-mcp --transport stdio
- Streamable HTTP (
--transport streamable-http): The primary recommended network transport (http://<host>:<port>/mcp). Provides fast, stateless, high-throughput streaming tool dispatch with native session cleanup. - Server-Sent Events (
--transport sse): Secondary transport (http://<host>:<port>/sse), supported for compatibility with legacy SSE-only MCP clients. - Standard I/O (
--transport stdio): For embedding directly inside local desktop agent processes.
Resources
The server exposes an informational MCP resource that agents can inspect:
config://app: Returns runtime server configuration (server name, debug mode, auth type, and active base URL).