Adding MCP support to PgDog
LLMs are pretty useful for debugging. Bugs that used to take us days to figure out can be resolved in a few minutes with a couple tailored prompts.
Giving your agent access to production telemetry however can be difficult. Most APMs, e.g., Datadog, Grafana, etc. are delayed by 15 seconds or more, and may not have all the context. During an incident, this could become a problem.
Our latest release of PgDog Enterprise (v2026-11-12) now ships with a built-in MCP server. It provides real time access to metrics, logs, configuration and system status, so your agent can quickly diagnose and resolve issues.
Getting started
If you’re using our Helm chart, you can enable MCP pretty easily:
# values.yaml
control:
config:
mcp:
enabled: true
This will launch an MCP server on the control plane’s /mcp endpoint. The server uses token-based authentication, which protects it from unauthorized access.
The server ships four tools:
| Tool | Description |
|---|---|
proxy_instances |
Up-to-date list of running PgDog containers and their status. |
proxy_metrics |
Fetch real time metrics for each configured database in pgdog.toml, direclty from the PgDog instances. |
proxy_logs |
Fetch real time PgDog container logs from Kubernetes. |
proxy_config |
Fetch latest configuration applied on all PgDog containers. |
All tools are read-only. Agents can’t make any changes to PgDog’s settings or deployment configuration.
They are also self-explanatory and an agent doesn’t require special instructions to use them. LLMs are able to figure out in which order to call each tool and how to use output of one tool as an input of another.
Enterprise edition
PgDog Enterprise is stable and publicly available. You can run it in your staging / local dev environments by pulling our Docker images directly from GitHub:
# PgDog
docker pull ghcr.io/pgdogdev/pgdog-enterprise:v2026-11-12
# Control plane
docker pull ghcr.io/pgdogdev/pgdog-enterprise/control:v2026-11-12
We provide Helm charts for both PgDog and its control plane, so installing it into your Kubernetes cluster is relatively simple.
If you need help or about to deploy PgDog to production, we’re always one call away.