Prometheus MCP
# Prometheus MCP
Proof-of-concept Prometheus MCP server.
## Prerequisites
Install `uv`: <https://docs.astral.sh/uv/getting-started/installation/>
Using `uv`, you can also install python.
## How to run
Clone this repo.
Update the .env file
```
uv add "mcp[cli]" pillow google-auth matplotlib requests python-dotenv
```
## Integrating with Claude
You can run the server with
```
uv --directory "/directory/to/prometheus-mcp" run server.py
```
So you may add this MCP server to your Claude MCP server configuration
```
{
"mcpServers": {
"Prometheus MCP": {
"command": "/path/to/uv",
"args": [
"--directory",
"/directory/to/prometheus-mcp",
"run",
"server.py"
]
}
}
}
```
See [MCP Quickstart](https://modelcontextprotocol.io/quickstart/server#testing-your-server-with-claude-for-desktop)
for more details for Claude specific instructions.
## Demo
[](https://github.com/etruong42/prometheus-mcp/raw/refs/heads/main/docs/prometheus-demo-video.mp4)
[Link to shared Claude chat in demo](https://claude.ai/share/c89412c1-5f19-430a-8c85-bf3c05033f81)
TDQS
Scored across 2 tools
The two tools target completely different aspects: one retrieves alert and recording rules, the other queries time series data and returns a plot. There is no overlap in functionality, making selection unambiguous.
Both tools share the 'prometheus_' prefix and follow snake_case. One uses a noun phrase ('alert_rules') and the other a verb-noun phrase ('query_range'), which is a minor inconsistency but still predictable and descriptive.
With only 2 tools for a complex system like Prometheus, the server is severely under-scoped. Many essential operations (e.g., listing metrics, instant queries, target discovery) are missing, limiting its usefulness.
The server covers only alert rules and range queries with plots, leaving out critical Prometheus capabilities such as instant queries, metric metadata, targets, or alert management beyond listing rules. Agents cannot perform basic monitoring workflows.