Firebase MCP Server
A general-purpose MCP server for Firebase that enables natural language interaction with Firestore, Auth, Storage, and Cloud Functions logs through schema-driven validation and context-efficient tools.
README
Firebase MCP Server
<p> <a href="https://www.npmjs.com/package/@dasasian/firebase-mcp-server"><img alt="npm" src="https://img.shields.io/npm/v/@dasasian/firebase-mcp-server?style=flat-square&color=235a9b"></a> <a href="LICENSE"><img alt="License: MIT" src="https://img.shields.io/badge/license-MIT-235a9b?style=flat-square"></a> <a href="https://modelcontextprotocol.io"><img alt="Model Context Protocol server" src="https://img.shields.io/badge/MCP-server-235a9b?style=flat-square"></a> <img alt="Node 18+" src="https://img.shields.io/badge/node-18%2B-5fa04e?style=flat-square"> </p>
General-purpose Model Context Protocol (MCP) server for Firebase (Firestore, Storage, Auth, Functions Logging) with schema-driven validation and context-efficient tools.
Features
- @ Mention Support - Reference Firestore documents with
@firebase:firestore://users/user-123 - Smart Autocomplete - MRU cache tracks accessed documents for quick re-reference
- Auto-Discovery - Shows both schema-based AND discovered collections
- Path-based schemas - Follows Firebase
firestore.rulesconvention - Schema evolution - Field status metadata (experimental → official → legacy)
- Hot reload - File watching, no restart needed when schemas change
- Flexible validation - Three modes: strict, warn (default), permissive
- Works without schemas - Discovery mode for exploring unknown databases
- Context-efficient tools - 99% token reduction for large datasets
- Index-aware queries - Validates queries against
firestore.indexes.json - Functions logging - SQL-like queries for Cloud Functions logs with aggregations and label filtering
Install
Add it to your MCP client (e.g. Claude Code) — runs via npx, no global install needed:
{
"mcpServers": {
"firebase": {
"command": "npx",
"args": ["-y", "@dasasian/firebase-mcp-server", "start", "./firestore-schemas.json"]
}
}
}
start takes an optional schema config path (default ./firestore-schemas.json) and an optional indexes path (default ./firestore.indexes.json) — see Schema Format. Firebase auth uses Application Default Credentials.
Or install the CLI globally:
npm install -g @dasasian/firebase-mcp-server
firebase-mcp start ./firestore-schemas.json
Development (from source)
npm install
npm run build
npm run cli -- start --config ./examples/basic/firestore-schemas.json
@ Mention Support (Resources)
Reference Firestore documents directly in Claude Code:
What's the email for @firebase:firestore://users/user-123?
Show me all posts: @firebase:firestore://posts/*
How It Works
When you type @firebase in Claude Code:
Shows schema-based collections (with validation):
- 📋 Users (User account documents)
- 📋 Posts (Blog post documents)
- 📋 Comments (Comments on posts)
Plus auto-discovered collections (no schema):
- 🔍 analytics
- 🔍 sessions
- 🔍 audit_logs
Configuration
# Enable/disable auto-discovery (default: true)
export FIRESTORE_AUTO_DISCOVER=true
# Cache duration in seconds (default: 300 = 5 minutes)
export FIRESTORE_DISCOVERY_CACHE_TTL=300
Auto-discovery cost: ~$0.0003/day (negligible)
Tools
The server ships 33 tools in four groups. Every tool declares MCP annotations
(readOnlyHint, destructiveHint, idempotentHint, openWorldHint) so a client
can auto-approve reads and ask before writes.
Choosing which tools to load
All 33 tool definitions cost roughly 12k tokens of context on every session. If you only need part of the surface, narrow it:
# Only Firestore (12 tools, ~4.3k tokens)
firebase-mcp start --tools firestore
# Firestore plus Auth
firebase-mcp start --tools firestore,auth
Or set it in your MCP client config:
export FIREBASE_MCP_TOOLS=firestore,storage
The --tools flag wins over the environment variable. Leaving both unset loads
everything, so upgrading never hides a tool you were already using. Calling a
tool from a group you switched off returns an error naming the group to add.
| Group | Tools | ~Tokens |
|---|---|---|
firestore |
12 | 4,341 |
storage |
14 | 4,152 |
auth |
6 | 2,075 |
logs |
1 | 1,817 |
| all (default) | 33 | 12,387 |
Firestore (12) — --tools firestore
firestore_show_collections— Show collections (read)firestore_read— Read document (read)firestore_export— Export collection (read)firestore_validate— Validate against schema (read)firestore_query_select— Query documents (read)firestore_query_collection_group— Query collection group (read)firestore_count— Count documents (read)firestore_sum— Sum a field (read)firestore_stats— Collection statistics (read)firestore_import— Import document (write, destructive)firestore_update— Update documents (write, destructive)firestore_delete— Delete documents (write, destructive)
Firebase Auth (6) — --tools auth
firebase_auth_list_users— List users (read)firebase_auth_get_user— Get user (read)firebase_auth_create_user— Create user (write)firebase_auth_update_user— Update user (write, destructive)firebase_auth_delete_user— Delete user (write, destructive)firebase_auth_revoke_sessions— Revoke sessions (write, destructive)
Firebase Storage (14) — --tools storage
firebase_storage_list_buckets— List buckets (read)firebase_storage_ls— List files (read)firebase_storage_stat— File metadata (read)firebase_storage_find— Find files (read)firebase_storage_get_url— Get file URL (read)firebase_storage_get_access— Get file access (read)firebase_storage_read— Download file to a local temp path (write)firebase_storage_upload— Upload file (write, destructive)firebase_storage_rm— Delete file (write, destructive)firebase_storage_cp— Copy file (write, destructive)firebase_storage_mv— Move file (write, destructive)firebase_storage_sync— Sync bucket to local (write, destructive)firebase_storage_push— Push local to bucket (write, destructive)firebase_storage_set_access— Set file access (write, destructive)
Cloud Logging (1) — --tools logs
firebase_functions_logs— Query Cloud Functions logs with SQL-like syntax (write)
firebase_storage_read and firebase_functions_logs are not marked read-only
because they write: the first downloads to a local temp file, the second updates
its auto-discovered logging schema on disk.
Reading function logs needs an extra IAM role. The Firebase Admin SDK service account has no Cloud Logging access by default, so
firebase_functions_logsfails withPERMISSION_DENIED: Permission denied for all log viewseven when every other tool works. Grant the role and allow a few minutes for it to take effect:gcloud projects add-iam-policy-binding PROJECT_ID \ --member="serviceAccount:firebase-adminsdk-xxxxx@PROJECT_ID.iam.gserviceaccount.com" \ --role="roles/logging.viewAccessor"
roles/logging.vieweralone may not be enough — the error names log views, and thelogging.views.accesspermission is inviewAccessor.
Example queries:
// Discover what functions exist
{"distinct": "functionName"}
// Show recent errors
{"where": [{"field": "severity", "operator": "==", "value": "ERROR"}], "limit": 20}
// Top error patterns with counts (use "message" — structured logs have no textPayload)
{"groupBy": ["message"], "aggregates": [{"field": "*", "operation": "count", "alias": "count"}], "where": [{"field": "severity", "operator": "==", "value": "ERROR"}], "orderBy": [{"field": "count", "direction": "desc"}], "limit": 10}
// Filter by custom labels (e.g., user, environment)
{"where": [{"field": "labels.user_id", "operator": "==", "value": "123"}]}
Schema Format
Schemas follow Firebase's path-based convention:
{
"schemas": {
"/organizations/{organizationId}": {
"description": "Organization documents",
"schema": {
"type": "object",
"required": ["id", "name"],
"properties": {
"id": { "type": "string" },
"name": { "type": "string" }
}
}
},
"/organizations/{organizationId}/products/{productId}": {
"description": "Product catalog",
"schema": {
"type": "object",
"properties": {
"name": { "type": "string" },
"category": {
"type": "string",
"x-status": "legacy",
"x-replacedBy": "productType"
},
"productType": {
"type": "string",
"x-status": "experimental"
}
}
},
"timestampFields": ["createdAt", "updatedAt"]
}
}
}
Documentation
- Installation Guide
- @ Mention Support Guide - How to use
@firebase:firestore://references - Configuration Reference
- Schema Creation Guide
- Functions Logging Guide - Query Cloud Functions logs with SQL-like syntax, aggregations, and custom labels
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。