mcpField
F
firebase-mcp-server

firebase-mcp-server

Local
dasasian.comCloud & Infra0Updated 10w ago

About firebase-mcp-server

firebase-mcp-server gives Firebase developers MCP tools for Firestore, Storage, Auth, and Cloud Functions logs. Schema validation, collection discovery, and selectable tool groups support both structured database work and exploration without schemas.

Key features

  1. 01Query and modify Firestore documents
  2. 02Manage Storage files
  3. 03Manage Firebase Auth users
  4. 04Validate schemas and discover collections
  5. 05Query Cloud Functions logs
  6. 06Select MCP tools by group

Use cases

  1. 01Validate Firestore data against schemas
  2. 02Explore collections in an unknown database
  3. 03Manage Firebase users and stored files
  4. 04Filter and aggregate function logs

MCP support

It exposes tools for Firestore, Storage, Auth, and logs to MCP clients, with selectable tool groups. MCP annotations describe tool read/write and risk properties.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"dasasian-firebase": {
"command": "npx",
"args": [
"-y",
"@dasasian/firebase-mcp-server"
]
}
}
}
Run commandnpx -y @dasasian/firebase-mcp-server

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
npm@dasasian/firebase-mcp-server1.1.0

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Firebase MCP Server

npm License: MIT Model Context Protocol server Node 18+

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.rules convention
  • 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.

GroupTools~Tokens
firestore124,341
storage144,152
auth62,075
logs11,817
all (default)3312,387
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_logs fails with PERMISSION_DENIED: Permission denied for all log views even 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.viewer alone may not be enough — the error names log views, and the logging.views.access permission is in viewAccessor.

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

License

MIT

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.