casper mcp
MCP server for Terraform and AWS

Casper MCP Server

A live, queryable view of your Terraform infrastructure for AI agents. Casper turns repos, state, drift, and policies into structured context your assistant can actually use.

Read docs
MCP
typed tools
0
write actions
1
live graph
claudecursorcodexcasper · mcp.tfawsstate
agentsliveinfrastructure

What it does

Infra context without giving agents write access.

Agent-ready context

  • Terraform-aware assistant workflows
  • Structured JSON responses
  • Session-level repo loading
  • Read-only cloud inspection

Infrastructure graph

  • Resources, modules, tags, and refs
  • HCL impact simulation
  • AWS state comparison
  • Policy violations surfaced inline

Team workflows

  • Review pull requests before apply
  • Find drift in production resources
  • Recommend existing modules
  • Block risky changes in CI

Core features

The tools an AI reviewer needs before it can reason about infra.

Find resources

Search by name, type, tag, or attribute across .tf and .tfstate.

Simulate changes

Parse proposed HCL, then get blast radius, broken refs, and policy hits before apply.

Detect drift

Compare Terraform state vs live AWS via read-only Describe APIs.

Enforce policies

Define org rules in .casper/policies.yaml. Violations surface inline.

Read-only by design

Casper never writes Terraform or AWS resources. render_graph writes one local HTML file; that's the only thing on disk.

FAQ

Common questions

Model Context Protocol is an open standard for connecting AI assistants to external systems. Instead of feeding your agent raw files, an MCP server exposes typed tools. Casper exposes a focused set of them for querying Terraform infrastructure — see the docs for the full list.

Early access

Built in the open, rough edges included.

v0 · early

Casper is in its early days. Expect discrepancies, missing provider coverage, and rough edges as the graph, drift, and policy layers mature. Feedback shapes what ships next.

If you want Casper as part of your real workflow, reach out and we'll build it together — issues you hit, providers you need, and conventions your team actually uses.

Put your infra graph in the agent loop.

Start with a read-only MCP server that works with Claude Code, Claude Desktop, Cursor, and any client that can call MCP tools.