1. Conduid
  2. RAG
  3. Mimic AI
MCP server · RAG

Mimic AI

Builds Figma from your design system: real components and tokens, enforced, learning every build.

Unclaimed docs
39Low

Scored yesterday · breakdown

About Mimic AI

Mimic AI is an MCP server in the RAG category: builds Figma from your design system: real components and tokens, enforced, learning every build.

The repository has 13 stars and 0 forks.

Install

Claude Code
claude mcp add mimic-ai -- npx -y @miapre/mimic-ai
npx
npx -y @miapre/mimic-ai

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Ask AI

✦

Ask AI about Mimic AI

Powered by Claude · Grounded in docs

I know everything about Mimic AI. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

  • ✓README presentThe repository ships documentation.

5 checks have not been run yet.

README

mimic

Composable persona prompts for any LLM — an MCP server.

mimic constructs structured LLM system prompts from reusable markdown fragments. Drop .md files into categorized directories, and mimic composes them into complete personas. Works across all MCP-compatible clients: Claude Code, Codex CLI, Gemini CLI, Cursor, VS Code, Windsurf, Zed, and more.

Ships with sensible defaults — plug and play out of the box.

Quick Start

Install:

cargo install mimic-ai

Add to Claude Code:

claude mcp add --transport stdio mimic -- mimic

Use:

> compose a backend engineer persona with go and postgresql skills

mimic's compose tool returns:

You are a senior backend engineer with deep experience building distributed systems,
designing APIs, and operating production services at scale...

## Expertise

You have expert-level proficiency in Go...

You have deep expertise in PostgreSQL...

How It Works

mimic organizes prompt fragments into five categories:

Category Purpose Example
Persona Core role identity backend-engineer, technical-writer
Skill Technical expertise go, postgresql, kubernetes
Context Situational framing code-review, debugging
Tone Communication style concise, pedagogical
Constraint Rules and boundaries no-frameworks, security-first

Fragments compose in fixed order: persona + skills + contexts + tones + constraints.

Fragment Format

Minimal markdown with optional YAML frontmatter:

---
description: Senior backend engineer focused on distributed systems
tags: [backend, distributed-systems, apis]
---
You are a senior backend engineer with deep experience building distributed systems,
designing APIs, and operating production services at scale.

Layered Resolution

Fragments resolve in priority order:

  1. Project-local (.mimic/) — project-specific overrides
  2. User global (~/.mimic/) — your personal defaults
  3. Built-in — ships with the binary

Override any built-in fragment by placing a file with the same name in your global or project-local directory.

MCP Interface

Tools

recommend — Get recommended fragments for a persona. Returns a flat, categorized list of skills, contexts, tones, and constraints that match the persona's skill groups. Use this before compose to see what's available.

Parameter Type Required Description
persona string yes Persona name (e.g. "backend-engineer")
groups string[] no Override the persona's skill groups (e.g. ["backend", "data"])
tags string[] no Filter recommendations by tags (e.g. ["security"])

compose — Build a system prompt from fragments.

Parameter Type Required Description
persona string yes Persona name (e.g. "backend-engineer")
skills string[] no Skill names (e.g. ["go", "postgresql"])
contexts string[] no Context names (e.g. ["code-review", "greenfield-project"])
tones string[] no Tone names (e.g. ["concise", "pedagogical"])
constraints string[] no Constraint names (e.g. ["no-frameworks"])

list — Browse available fragments. Returns JSON with name, category, description, and tags. Optionally filter by category, tag, or group.

check_update — Check if a newer version of mimic is available. Returns JSON with current, latest, and update_available fields. Results are cached for 1 hour.

Note: When check_update has been called and an update is available, compose will automatically append an update notice to its output.

resolve — Advanced: returns a raw graph of fragment relationships. Most clients should use recommend instead. Given a persona, tags, or groups, returns nodes and edges connected by shared tags, groups, and skill groups.

Parameter Type Required Description
persona string no Starting persona to anchor the graph
tags string[] no Seed tags — only fragments sharing at least one tag are included (tones always included)
groups string[] no Filter skills to these groups (overrides persona's skill_groups)
include_edges boolean no Include edges in the result (default true)

Resources

Browse fragments individually via mimic://fragments/{category}/{name} URIs.

Prompts

Common compositions exposed as MCP prompts (e.g. /mimic-backend-engineer).

Completions

Autocomplete support for all tool parameters — any client that supports MCP completions gets typeahead for free.

Editor Setup

Claude Code

claude mcp add --transport stdio mimic -- mimic

Cursor / VS Code

Add to .cursor/mcp.json or VS Code MCP settings:

{
  "mcpServers": {
    "mimic": {
      "command": "mimic",
      "args": [],
      "transportType": "stdio"
    }
  }
}

Codex CLI

codex mcp add mimic -- mimic

See the full documentation for more editor setup guides.

Custom Fragments

Create your own fragments in ~/.mimic/ (global) or .mimic/ (project-local):

mkdir -p ~/.mimic/skills
cat > ~/.mimic/skills/elixir.md << 'EOF'
---
description: Elixir and OTP expertise
tags: [language, elixir, beam]
---
You have deep expertise in Elixir and the OTP framework. You write idiomatic
Elixir using GenServers, Supervisors, and the actor model. You understand BEAM
VM characteristics and can reason about fault tolerance and concurrency.
EOF

Linting

Validate your fragment files for correctness:

mimic lint              # show errors only
mimic lint --warnings   # show errors and warnings

The linter checks built-in, global (~/.mimic/), and project-local (.mimic/) fragments. Exit code is 1 if any errors are found.

Built-in Fragments

mimic ships with 164 built-in fragments across all five categories.

Personas (43): backend-engineer, data-engineer, data-scientist, designer, devops-engineer, engineering-manager, frontend-engineer, fullstack-engineer, ml-engineer, mobile-engineer, platform-engineer, product-manager, qa-engineer, security-engineer, solutions-architect, sre-engineer, technical-writer, staff-backend-engineer, staff-data-engineer, staff-designer, staff-devops-engineer, staff-engineering-manager, staff-frontend-engineer, staff-ml-engineer, staff-mobile-engineer, staff-platform-engineer, staff-product-manager, staff-qa-engineer, staff-security-engineer, staff-technical-writer, principal-backend-engineer, principal-data-engineer, principal-designer, principal-devops-engineer, principal-engineering-manager, principal-frontend-engineer, principal-ml-engineer, principal-mobile-engineer, principal-platform-engineer, principal-product-manager, principal-qa-engineer, principal-security-engineer, principal-technical-writer

Skills (80): agile, ai-ml, android, angular, ansible, api-design, appsec, aws, azure, c, ci-cd, clickhouse, cloud-security, cpp, cryptography, csharp, css, datamodeling, deno, devsecops, docker, dynamodb, elasticsearch, elixir, event-driven, flutter, gcp, git, github-actions, go, graphql, grpc, html, htmx, identity, ios, java, kafka, kotlin, kotlin-mobile, kubernetes, linux, llm, microservices, mongodb, mysql, networking, nextjs, nginx, nix, observability, performance, php, postgresql, python, rabbitmq, react, react-native, redis, regex, rest-api, ruby, rust, scala, serverless, shell-scripting, spark, sql, sqlite, svelte, swift, system-design, tailwind, technical-writing, terraform, testing, typescript, vite, vue, webassembly

Contexts (18): architecture-review, code-review, debugging, documentation, estimation, greenfield-project, incident-response, legacy-codebase, mentoring, migration, onboarding, open-source, pair-programming, performance-tuning, post-mortem, prototyping, refactoring, security-audit

Tones (8): casual, concise, dry-humor, empathetic, formal, opinionated, pedagogical, socratic

Constraints (15): accessibility-first, backwards-compatible, compliance, cost-conscious, deterministic, memory-constrained, minimal-downtime, no-frameworks, offline-first, open-source-only, performance-critical, privacy-first, security-first, test-driven, type-safe

Contributing

See CONTRIBUTING.md. The easiest way to contribute is adding new fragment files — no Rust knowledge required.

License

MIT

README copy captured when this server was indexed; it may lag the repository. The original is authoritative.

Questions

About Mimic AI

How do I install Mimic AI?

Run claude mcp add mimic-ai -- npx -y @miapre/mimic-ai, then add the server to your MCP client's configuration.

Is Mimic AI safe to use with an AI agent?

Its trust score is 39 out of 100 (low). It passes 1 of the 1 static security checks run so far. It has no ConduID identity yet, so agent calls to it are not receipted.

Is Mimic AI still maintained?

Conduid hasn't recorded a commit date for this repository yet. Check the repository directly for recent activity.