npm — FAI MCP Server

Add AI Infra, Platform, and App Knowledge to Your Agent

Quick Run

npx frootai-mcp

Install Global

npm i -g frootai-mcp

npm Registry

Open on npm Website →

Without FrootAI MCP

Agent searches the internet — slow, noisy
Burns 5,000+ tokens per architecture query
May hallucinate design guidance
Generic answers — no Azure patterns

With FrootAI MCP

Queries curated 664KB knowledge base
90% less token burn
Zero hallucination — grounded in verified docs
Azure-specific best practices & patterns

Connect to Your Client

Claude Desktop

claude_desktop_config.json

{"mcpServers":{"frootai":{"command":"npx","args":["frootai-mcp"]}}}

VS Code / Copilot

.vscode/mcp.json

{"servers":{"frootai":{"command":"npx","args":["frootai-mcp"]}}}

Microsoft Foundry

Agent → Tools → Add MCP

Point to npx frootai-mcp

Cursor

MCP settings

{"mcpServers":{"frootai":{"command":"npx","args":["frootai-mcp"]}}}

Windsurf

MCP settings

{"mcpServers":{"frootai":{"command":"npx","args":["frootai-mcp"]}}}

Copilot Studio

Copilot Studio tools

Add MCP connector

Gemini / Codex

MCP config

{"mcpServers":{"frootai":{"command":"npx","args":["frootai-mcp"]}}}

Tools Agent Receives

Static + live + agent chain + ecosystem + FAI engine + scaffold + marketplace = full MCP toolkit

list_modules

Browse knowledge modules by FROOT layer

get_module

Read any module content (F1–T3)

lookup_term

Comprehensive AI/ML term definitions

search_knowledge

Full-text search all modules

get_architecture_pattern

7 pre-built decision guides

get_froot_overview

Complete FROOT framework summary

Live Tools (v2 — network-enabled)

fetch_azure_docs

Search Microsoft Learn for Azure docs

fetch_external_mcp

Find MCP servers from registries

list_community_plays

List solution plays from GitHub

get_github_agentic_os

.github 7 primitives guide

Agent Chain Tools (Build → Review → Tune)

agent_build

Play-aware architecture, code patterns, pinned deps (v6.3+)

agent_review

12-rule code engine: severity + file:line + suggestion (v6.3+)

agent_tune

Validates openai.json/guardrails/play configs against thresholds (v6.3+)

AI Ecosystem Tools (v3)

get_model_catalog

Compare GPT-4o, Claude, Llama, Phi models

get_azure_pricing

Token pricing for Azure OpenAI models

estimate_cost

Estimate monthly cost for a solution play

compare_models

Side-by-side model comparison

compare_plays

Compare two solution plays

semantic_search_plays

Natural language play search

embedding_playground

Compare texts for semantic similarity — v6.7+ real Azure OpenAI 1536-dim embeddings (auto when AZURE_OPENAI_ENDPOINT set) or char-ngram cosine fallback

Developer Tools

validate_config

Validate TuneKit configs (openai.json, guardrails.json) against best practices

generate_architecture_diagram

Generate Mermaid.js architecture diagram for any solution play

run_evaluation

Quality scoring — groundedness, relevance, coherence, fluency (pass/fail)

FAI Engine — The UniFAIng Glue (FAI Protocol Runtime)

wire_play

Resolve FAI Protocol context, bind primitives, activate guardrails

inspect_wiring

Audit a play's wiring — primitives, context, health score

validate_manifest

Validate fai-manifest.json schema + completeness

get_play_detail

Full play spec — services, config, WAF alignment, evaluation

list_primitives

Browse primitives by type (agents, skills, hooks…)

evaluate_quality

Run quality evaluation with configurable pass/fail thresholds

run_eval_live

POST cases to a live endpoint, score with simple/LLM-as-judge, aggregate mean/p95 vs threshold (v6.6+)

Scaffold & Create — Generate Production-Ready Plays

scaffold_play

Generate a complete play (24+ files) with FAI Protocol auto-wired

create_primitive

Create an agent, instruction, or skill with proper frontmatter

smart_scaffold

Semantic search → find best matching play → scaffold it

scaffold_component

Copy-paste-ready code for a single Azure AI building block (15 components, py/ts/cs) (v6.4+)

get_play_config

Starter configs for a play: openai/guardrails + play-specific (search.json, speech.json, etc.) (v6.4+)

get_dependencies

Pinned dependency lists per language (python/typescript/csharp) for a play (v6.4+)

generate_bicep

Play-specific Bicep IaC with Managed Identity, RBAC, tags, optional private endpoints (v6.5+)

Workspace — Detect, Inspect, Recommend

analyze_workspace

Inspect cwd — detected play, primitives, configs, Bicep services, FAI manifest, gaps + recommendations

Marketplace — npm for AI Primitives (77+ Plugins)

marketplace_search

Semantic search across plugins by use case

marketplace_browse

Paginated listing with 8 category filters

install_plugin

Copy plugin primitives into .github/ with conflict detection

uninstall_plugin

Remove plugin primitives cleanly

list_installed

Scan .github/ and match against marketplace registry

check_compatibility

Validate plugin + play alignment + WAF + file conflicts

validate_plugin

Check plugin.json schema, naming, and file refs

compose_plugins

Multi-install with cross-plugin conflict detection

publish_plugin

Validate → generate marketplace entry → register

check_plugin_updates

Compare installed vs marketplace versions

resolve_dependencies

Topological dependency resolution + install order

list_external_plugins

Browse community plugins from external sources

marketplace_stats

Full analytics: totals, categories, top plugins, play coverage

See It In Action

Real conversations with the FAI MCP server — what you ask and what you get back

🔍

You ask: "I want to build a RAG chatbot"

semantic_search_plays({ query: "RAG chatbot with Azure AI Search" })

Result: Play 01 (Enterprise RAG, 92% match), Play 21 (Agentic RAG, 78%), Play 09 (AI Search Portal, 65%)

Instead of Googling, your AI agent instantly finds the best pre-built solution from all plays — ranked by BM25 relevance.

🏗️

You ask: "Scaffold a new project from scratch"

scaffold_play({ name: "Invoice Processor", model: "gpt-4o" })

Result: 24 files created — agents, skills, config, infra, evaluation — all auto-wired via FAI Protocol

One tool call creates a complete, production-ready project. No other MCP server does this.

🔌

You ask: "Wire my play and check if everything connects"

wire_play({ playId: "01" })

Result: ✅ All primitives wired: 1 agent → 3 instructions → 3 skills → 3 hooks, shared context (4 knowledge modules + 5 WAF pillars), 43ms

"Wiring" = the FAI Engine reads your manifest, resolves all paths, injects shared context, and verifies quality gates. Like docker compose up for AI primitives.

📊

You ask: "Check if my AI responses meet quality standards"

evaluate_quality({ scores: { groundedness: 0.97, relevance: 0.88, safety: 0 } })

Result: ✅ PASS — Groundedness 0.97 ≥ 0.95, Relevance 0.88 ≥ 0.85, Safety 0 = 0. Each play has its own quality bar.

Guardrails are defined per-play in fai-manifest.json. Fail a threshold → the engine tells you which metric to fix.

⚡

You ask: "Build → Review → Tune workflow"

agent_build → agent_review → agent_tune("Invoice processing API with Azure Document Intelligence")

Result: Build: architecture guidance → Review: security + quality audit → Tune: production readiness verdict

Mirrors how real teams work: architect → code reviewer → QA. Three agents forming a quality chain.

What Makes This Different

Other MCP servers wrap APIs. FrootAI MCP is a domain knowledge engine with a protocol that binds AI primitives together.

Feature	GitHub MCP	Anthropic MCP	FAI MCP
Runtime engine	No	No	✅ FAI Engine
Wires primitives	No	No	✅ fai-manifest.json
Quality gates	No	No	✅ Guardrails per play
Scaffolds projects	No	No	✅ 24 files, auto-wired
Knowledge base	No	No	✅ Modules, BM25
Plugin marketplace	No	No	✅ 77+ plugins
Progress tokens	No	No	✅ Per-file progress
Tool annotations	Partial	No	✅ All tools

Setup Guide VS Code Extension Ecosystem Back to FrootAI