Skip to main content
Glama
Sign In
tfplan-mcp

tfplan-mcp

by 0xmortuex
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

tfplan-mcp the Hard Way

Build an MCP server that lets an AI risk-review a Terraform / OpenTofu plan, from scratch, in seven small modules. You end with a working miniature of tfplan-mcp: it reads a plan file, summarizes it, ranks the dangerous changes, diffs one resource with secrets masked, and compares two plans, all over MCP stdio, and it never runs terraform.

Every module has:

  • README.md - the why, in plain prose

  • code you can read in one sitting

  • main.py - runs the code and asserts the result, so the claims in the README are checked

  • break-it.md - challenges that make you find the sharp edges yourself

verify.py runs every main.py for real. CI does the same on Ubuntu and Windows.

Run it

python -m venv .venv
.venv/Scripts/activate        # Linux/macOS: source .venv/bin/activate
pip install -r requirements.txt
python verify.py              # all seven modules
python verify.py 03 05        # just these
python 03-risk-heuristics/main.py

Only module 07 needs the mcp package. Modules 01-06 are plain Python.

Related MCP server: tfminder

Modules

#

Module

You learn

01

The plan JSON

what terraform show -json contains, and refusing things that are not plans (a state file looks almost identical)

02

Action semantics

create/update/delete/replace, why a replace is two actions and the order matters, no-ops that are really moves and imports

03

Risk heuristics

ranking changes: stateful types, new exposure only (compare before and after), tags-only demotion

04

Diffing nested values

walking before/after/after_unknown together, paths like ingress[1].cidr_blocks[0], (known after apply)

05

Masking secrets

two independent layers: Terraform's sensitivity masks and name heuristics

06

Comparing plans

new / removed / changed action / changed values between two plans

07

The MCP wrapper

tools, a directory sandbox, errors an agent can read, and a plant-a-secret test over real stdio

The sample plans in plans/ are small hand-written terraform show -json documents (format_version 1.2). The fake secret in them is TUTORIAL-hunter2; the tests search every output for it.

MIT licensed.

This server cannot be deployed

Claude Code, inside the Godot 4 editorFlockbay puts an AI agent in the real Godot 4 editor: it writes GDScript, runs your game and fixes what breaks. Free, or bring your own Claude Code or Codex.flockbay.comAd

Maintenance

ActivityMaintained
ResponsivenessNo recent community issues

Resources

Related MCP Connectors

Related MCP Servers

  • iac-copilot

    togleid
    A
    license
    A
    quality
    C
    maintenance
    Enables AI coding agents to search a security corpus, scan Terraform plan JSON for cited findings, and list sources via MCP tools grounded in CIS AWS Foundations and OWASP Top 10 CI/CD risks.
    3
    Apache 2.0

  • tfminder

    MarckMorris
    A
    license
    Not graded
    quality
    C
    maintenance
    Lets AI agents propose and review Terraform plans through MCP while requiring human approval before any exact reviewed plan is applied, with policy checks and a tamper-evident audit log.
    Apache 2.0

  • terraform-mcp

    imashish-in
    A
    license
    Not graded
    quality
    C
    maintenance
    Enables state-aware Terraform/OpenTofu plan evaluation against live AWS topology and CloudWatch telemetry, detecting reliability blast radius and FinOps waste before merge.
    Apache 2.0

  • Cloud DevOps MCP Server

    alexcgodwin
    A
    license
    A
    quality
    A
    maintenance
    Enables AI clients to assess infrastructure-as-code deployment risk, generate incident response runbooks, review CI/CD delivery readiness, and calculate SLO error budgets through structured advisory tools. Runs locally over stdio without cloud credentials or external API calls, returning guidance for platform work, pull request review, and production readiness checks.
    12
    185 npm
    MIT