queryLABSCO SUMMARY
Domain questions answered from your documents, with the sources shown.
The part worth spending time on is not the install but the description you write for the agent: that is what decides when your client reaches for it instead of answering from memory. If running one is not appealing, the platform hosts a server whose query tool behaves the same way.
What it is
A server that sits between an MCP client and a Contextual AI agent: the question goes to the agent you nominate, the agent searches your datastore, and the answer comes back grounded in what it found.
What you get
- Answers drawn from your knowledge base rather than from model training
- Citations and attributions alongside each answer
- Conversation context carried across turns, so a follow-up does not need the question restated
- A hosted server inside the platform, if you would rather not run one
- A server you configure yourself, including the description that tells the client when a question belongs to it
Requirements
Python 3.10 or higher and a Contextual AI API key. The server hosted inside the platform needs neither — connect to it and use the tools it offers.
Setup effort
Build from source — git clone https://github.com/ContextualAI/contextual-mcp-server.git
USE CASES
01
Asking about material the model has never seen
Send the question to an agent whose datastore holds your own documents.
02
Checking where an answer came from
Citations and attributions come back with the response, so a claim can be traced to a document.
03
Following up without repeating yourself
Conversation context is kept across turns.
04
Skipping local setup entirely
Connect to the server hosted in the platform and call its query tool.
1 tool
Listed in project docs. Your client may show a different list.
BEFORE YOU INSTALL
What to know first
Tools you get
1 tool named in the project's documentation.
Requires
- Python 3.10
- API_KEY in your environment
- AGENT_ID in your environment
- uv
Where it runs
Both documented clients run the server on your machine.
INSTALLATION
Connect your client
This one is built from source rather than installed with a single command.
Server setup
Setup
TERMINAL
git clone https://github.com/ContextualAI/contextual-mcp-server.git
cd contextual-mcp-server
python -m venv .venv
source .venv/bin/activate # On Windows, use .venv\Scripts\activate
pip install -e .Set API_KEY and AGENT_ID (e.g. in a .env file)
Find the full path of uv with which uv — it goes where the configuration says $UV_PATH
Write mcp.json with the mcpServers block and move it to .cursor/mcp.json (or the Claude Desktop config directory)
Hosted endpoint
ENDPOINT URL
https://mcp.app.contextual.ai/mcp/Add this hosted endpoint to your MCP client using the credentials listed under Requirements.
MCP Clients
Cursor
CURSOR · JSON
{
"mcpServers": {
"ContextualAI-TechDocs": {
"command": "$UV_PATH",
"args": [
"--directory",
"${workspaceFolder}",
"run",
"multi-agent/server.py"
]
}
}
}A JSON file does not expand "$UV_PATH" (the README sets UV_PATH=$(which uv) in a shell), so the command has to be the full path that which uv prints. Cursor fills ${workspaceFolder} with the folder you have open, so open your contextual-mcp-server clone, or write its absolute path there instead.
The configuration on this page does not set API_KEY and AGENT_ID; the project's instructions put them in a .env file. Check which folder the server reads that file from before you rely on it.
Claude Desktop
CLAUDE DESKTOP · JSON
{
"mcpServers": {
"ContextualAI-TechDocs": {
"command": "$UV_PATH",
"args": [
"--directory",
"${workspaceFolder}",
"run",
"multi-agent/server.py"
]
}
}
}Claude Desktop expands neither value: "$UV_PATH" has to be the full path that which uv prints, and ${workspaceFolder} the absolute path of your contextual-mcp-server clone.
The configuration on this page does not set API_KEY and AGENT_ID; the project's instructions put them in a .env file. Check which folder the server reads that file from before you rely on it.
