[ PLAN REVIEW MCP ]

REVIEW THE PLAN.
NOT JUST THE PR.

A remote MCP server that checks your AI agent's implementation plan against your codebase's conventions — before a line is written. Works with Claude Code, Cursor, Codex, and any MCP-capable agent.

POST https://api.gradus.maiife.ai/mcp/vibecop · streamable HTTP · Bearer auth

[ STEP 0 ]

GET YOUR TOKEN

Mint an mcp-scoped token in VibeCop Settings → "Plan Review MCP · Agent Auth" — available on every plan. The raw token is shown once, so copy it immediately. Tokens expire after 90 days and can be revoked anytime.

Open Settings →

[ INSTALL ]

CONNECT YOUR AGENT

Claude Code CLI

claude mcp add --transport http vibecop https://api.gradus.maiife.ai/mcp/vibecop --header "Authorization: Bearer <TOKEN>"

.mcp.json

{
  "mcpServers": {
    "vibecop": {
      "url": "https://api.gradus.maiife.ai/mcp/vibecop",
      "headers": {
        "Authorization": "Bearer <TOKEN>"
      }
    }
  }
}

Replace <TOKEN> with the token from Step 0.

[ THE SKILLS ]

NINE SKILLS. ONE REGISTRY.

full-review10¢

"Review this PRD/spec against our codebase — the default; run on every plan before writing code."

reuseadrprofileimpactbreakingschemasimplicitydependencysecurity

reuse-check3¢

"Does this already exist?"

reuse

architecture-fit3¢

"Does this plan fit how our codebase is built?"

adrprofile

impact-analysis3¢

"What breaks if I touch these files?"

impact

schema-review3¢

"Is this migration safe?"

schema

breaking-change-review3¢

"Who consumes what I'm about to change?"

breaking

simplicity-review3¢

"Am I over-engineering this?"

simplicity

dependency-review3¢

"Should I add this package?"

dependency

security-review3¢

"What's the threat surface of this plan?"

security

test-plan-review(PLANNED)

"Check whether the plan adequately specifies test coverage for its proposed changes."

adr-draft(PLANNED)

"Draft an ADR capturing an approved plan's architectural decisions."

naming-review(PLANNED)

"Check whether proposed names match the codebase's established naming conventions."

[ THE TOOLS ]

SIX TOOLS.

list_skillsList available plan-review skills plus planned ones.

get_skillFull markdown instructions for a named skill.

get_project_contextStack profile, ADR titles, module directories, and the skill registry for a project.

vibecop_checkRun a skill against a submitted plan + extracted intents; returns verdict, findings, recommendations.

get_plan_reviewFetch a persisted plan review by id.

resolve_projectLook up projects by name; returns candidates when ambiguous, never silently picks one.

[ AGENT RULES ]

AUTOMATE THE HABIT

Drop this into your CLAUDE.md or .cursorrules so the agent runs a plan review automatically.

Agent rules

Before writing any code against a PRD or spec, use the "vibecop" MCP server:
1. Call list_skills to see available plan-review skills.
2. Call get_skill({ name: "full-review" }) and follow its instructions to extract intents
   (proposed utilities, modules, dependencies, endpoints, schema changes) from the document.
3. Call vibecop_check({ skill: "full-review", docTitle, docText, intents }) and read the
   verdict, findings, and recommendations before starting implementation.
If the verdict is "fail", resolve the findings or get explicit sign-off before proceeding.

[ INITIALIZE ]

PUT A REVIEWER IN THE LOOP
BEFORE THE CODE EXISTS.

GET STARTED FREE →VIEW PRICING →

REVIEW THE PLAN.NOT JUST THE PR.