DrDroid

droidctx

live infra context for your AI.

A single command that snapshots your services, dashboards, deploys, and alerts, then pipes it into Claude Code, Cursor, or any MCP-compatible agent. Your AI stops hallucinating about your stack and starts answering with real data.

Install droidctx View on GitHub
~/ops/orders-svc
$ curl -fsSL https://raw.githubusercontent.com/DrDroidLab/context-builder/main/install.sh | bash ==> Installed droidctx 0.7.2 (39 files, 18.4 MB) $ droidctx connect ✓ Grafana api.grafana.local · 12 dashboards ✓ Kubernetes prod · 6 ns · 142 pods ✓ Datadog via $DD_API_KEY · 38 services ✓ GitHub org: acme · 24 repos $ claude "why is orders-svc 5xx?" ▸ context: droidctx · 4 sources · refreshed 2s ago ▸ orders-svc on ECS us-east-1 · Redis prod-cluster p99 ↑ ▸ Root cause: connection pool exhausted since 14:32 $
THE PROBLEM

Production is too vast to guess.

A typical stack spans Kubernetes, ECS, Lambda, with metrics in Grafana, alerts in Datadog, schemas in PostgreSQL. No agent can navigate this without a map. Without one, it hallucinates its way through 15 tool calls and gives up.

Without droidctx
> Debug why payment latency spiked Checking Kubernetes pods... (service is on ECS) Writing PromQL from scratch... (dashboard exists in Grafana) Found redis-test-instance... (wrong Redis cluster) Checking CloudWatch... (team uses Datadog) Agent gave up after 15 tool calls.
With droidctx
> Debug why payment latency spiked payment-service → ECS us-east-1 (from context) Opening "Payment Overview" in Grafana Redis prod-cluster p99 spike at 14:32 Connection pool exhausted — 512 / 512 Root cause found in 4 tool calls.
5–10

Observability tools per team

Each with its own dashboards, alerts, and naming conventions your agent doesn't know.

50+

Dashboards to navigate

Without a map, agents check the wrong tool every time.

0

Known to your agent by default

Without context files, agents guess. droidctx sync changes that.

INSTALL

One command. Three ways to install.

Pick your preferred method. All options install the same droidctx binary with an isolated Python environment.

$ curl -fsSL https://raw.githubusercontent.com/DrDroidLab/context-builder/main/install.sh | bash
HOW IT WORKS

Auto-detect. Snapshot. Serve.

droidctx is three things in one binary: a connector layer, a context builder, and an MCP server. Each step is independently scriptable.

01 / DETECT

Sees what's already configured.

Reads ~/.kube/config, env vars, ~/.aws/credentials, and your local Grafana datasources. Zero config for ~80% of stacks.

AUTO-DISCOVERY · 25+ SOURCES
02 / SNAPSHOT

Builds context on every prompt.

When your AI asks, droidctx pulls live data: current deploys, dashboards, recent alerts, topology. Cached for 30s, then refreshed. Always fresh, never stale.

CACHED · REFRESHED EVERY 30S
03 / SERVE

Speaks MCP. And REST. And stdio.

One binary serves Model Context Protocol for Claude Code, a REST API for custom tools, and stdio for shell pipelines. Same data, three doors.

MCP · REST · STDIO
WHAT GETS GENERATED

Structured context your agent can navigate.

A hierarchy of markdown files covering services, dashboards, schemas, alerts, and cross-service mappings, generated from your live infrastructure, not hand-written docs that go stale overnight.

droidctx-context

Click files to preview. Generated from your live infrastructure.

CONNECTORS

25+ tools, one command.

Read-only access. No write permissions, no agents to deploy, no vendor lock-in.

Grafana Grafana
Datadog Datadog
New Relic New Relic
Kubernetes Kubernetes
PostgreSQL PostgreSQL
MongoDB MongoDB
Elasticsearch Elasticsearch
AWS AWS
GitHub GitHub
Sentry Sentry
Jenkins Jenkins
ArgoCD ArgoCD
CloudWatch CloudWatch
Jira Jira
PostHog PostHog
SigNoz SigNoz
GKE GKE
Google Cloud Google Cloud
ClickHouse ClickHouse
Prometheus Prometheus
Victoria Metrics Victoria Metrics
Grafana Loki Grafana Loki
OpenSearch OpenSearch
Coralogix Coralogix
Azure Azure
25+ connectors · Cloud, Kubernetes, observability, databases, CI/CD Read-only · MIT licensed
MADE FOR CLAUDE CODE

Drop into any MCP-aware agent.

droidctx implements the Model Context Protocol, so it slots into Claude Code, Cursor, Zed, or your homegrown agent with one config line.

  • Claude Code, automatic discovery via claude_desktop_config.json
  • Cursor & Zed, paste the MCP URL into settings, done.
  • Custom agents, REST/SSE endpoints for anything that speaks HTTP.
  • Shell pipelines, stdio mode for jq, scripts, and CI jobs.
claude_desktop_config.json
{   "mcpServers": {     "droidctx": {       "command": "droidctx",       "args": ["serve", "--mcp", "--stdio"],       "env": {         "DD_API_KEY": "$DD_API_KEY"       }     }   } } # Restart Claude. droidctx is now a tool.
YOUR PARTNER IN RELIABILITY

One binary. Smarter agent.

Install droidctx and your AI knows your stack within thirty seconds. No accounts, no data shared, no waiting.

Install droidctx Source on GitHub
FAQ

Common questions.