db-mcp-server MCP — Cas d'usage, Installer & Démo en direct

Pourquoi l'utiliser

Fonctionnalités clés

Multi-DB in one server — switch without restarting
Per-connection auto-generated tools (e.g. query_prod, schema_analytics)
TimescaleDB-native tools for time-series ops
Oracle: 10g-23c, RAC, Cloud Wallet

Démo en direct

Aperçu en pratique

db.replay ▶ prêt

0/0

Installer

Choisissez votre client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "db": {
      "command": "TODO",
      "args": [
        "See README: https://github.com/FreePeak/db-mcp-server"
      ],
      "_inferred": true
    }
  }
}

Ouvrez Claude Desktop → Settings → Developer → Edit Config. Redémarrez après avoir enregistré.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "db": {
      "command": "TODO",
      "args": [
        "See README: https://github.com/FreePeak/db-mcp-server"
      ],
      "_inferred": true
    }
  }
}

Cursor utilise le même schéma mcpServers que Claude Desktop. La config projet l'emporte sur la globale.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "db": {
      "command": "TODO",
      "args": [
        "See README: https://github.com/FreePeak/db-mcp-server"
      ],
      "_inferred": true
    }
  }
}

Cliquez sur l'icône MCP Servers dans la barre latérale Cline, puis "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "db": {
      "command": "TODO",
      "args": [
        "See README: https://github.com/FreePeak/db-mcp-server"
      ],
      "_inferred": true
    }
  }
}

Même format que Claude Desktop. Redémarrez Windsurf pour appliquer.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "db",
      "command": "TODO",
      "args": [
        "See README: https://github.com/FreePeak/db-mcp-server"
      ]
    }
  ]
}

Continue utilise un tableau d'objets serveur plutôt qu'une map.

~/.config/zed/settings.json

{
  "context_servers": {
    "db": {
      "command": {
        "path": "TODO",
        "args": [
          "See README: https://github.com/FreePeak/db-mcp-server"
        ]
      }
    }
  }
}

Ajoutez dans context_servers. Zed recharge à chaud à la sauvegarde.

claude mcp add db -- TODO 'See README: https://github.com/FreePeak/db-mcp-server'

Une seule ligne. Vérifiez avec claude mcp list. Supprimez avec claude mcp remove.

Cas d'usage

Usages concrets : db-mcp-server

Run cross-database analysis from Claude

👤 Data engineers ⏱ ~25 min intermediate

Quand l'utiliser : You need to pull from Postgres (app) AND MySQL (legacy) in one conversation without switching tools.

Prérequis

config.json with both connections — Repo docs show the shape; store credentials in env refs, not inline

Déroulement

Start the server

Run ./bin/server -t sse -c config.json and confirm both connections come up.✓ Copié

→ Server logs 2 connections ok
Query each

From prod (Postgres): users signed up last week. From legacy (MySQL): orders attributed to them. Join in memory.✓ Copié

→ Combined dataset

Résultat : Cross-system insights without a data warehouse.

Pièges

Read-only assumption breaks — agent runs INSERTs — Use DB-level read-only users per connection; don't rely on agent discipline

Combiner avec : google-sheets

Manage TimescaleDB hypertables from Claude

👤 Observability / IoT teams ⏱ ~20 min advanced

Quand l'utiliser : You want to create/inspect hypertables and continuous aggregates without memorizing Timescale DDL.

Déroulement

Inspect existing hypertables

List all hypertables in metrics DB with chunk interval and row count.✓ Copié

→ Table of hypertables
Create a continuous aggregate

Create a 1-hour rollup on sensor_readings grouping by device_id, avg + max + min.✓ Copié

→ CAgg created; refresh policy configured

Résultat : Timescale ops in minutes, not Google searches.

Explain an unfamiliar schema to get onboarded fast

👤 Engineers inheriting a database ⏱ ~30 min beginner

Quand l'utiliser : Day 1 at a new team; you need a map of the DB.

Déroulement

Dump schema

Use schema_<conn_id> on prod. Return tables, FK graph, and row-count order.✓ Copié

→ Schema + FK map
Generate a glossary

For each table, infer a 1-line description from column names and sample rows (limit 5 per table).✓ Copié

→ Onboarding cheat sheet

Résultat : Working mental model in 30 min.

Combinaisons

Associez-le à d'autres MCPs pour un effet X10

db + google-sheets

Export a query result into a shared Sheet for non-tech stakeholders

Query top 50 customers by LTV from prod; write to the 'Top LTV' sheet.✓ Copié

db + prometheus

Cross-check DB slow-query findings with DB-level Prometheus metrics

For the slow query found via performance_prod, show the pg_stat_statements metrics from Prometheus for the same window.✓ Copié

Outils

Ce que ce MCP expose

Outil	Entrées	Quand appeler	Coût
query_<db_id>	sql: str (SELECT)	Read data	1 query
execute_<db_id>	sql: str (DDL/DML)	Writes — guarded by DB permissions	1 query
transaction_<db_id>	statements: str[]	Multi-statement atomic changes	1 tx
schema_<db_id>	table?: str	Discovery / onboarding	metadata query
generate_schema_<db_id>	format: sql\|json	Export for docs/version control	metadata queries
performance_<db_id>	sql?: str	Tune a slow query	plan + stats

Coût et limites

Coût d'exécution

Quota d'API: Your DBs' capacity
Tokens par appel: Large result sets burn tokens fast — LIMIT aggressively
Monétaire: Free MCP; DB hosting costs are yours
Astuce: Always add LIMIT / cap output; stream to a file via filesystem MCP for bigger pulls

Sécurité

Permissions, secrets, portée

Portées minimales : DB-level read-only user recommended for exploration

Stockage des identifiants : config.json references env vars; never commit with inline passwords

Sortie de données : Only to configured DB hosts

Ne jamais accorder : DB superuser to the MCP connection unless absolutely needed

Dépannage

Erreurs courantes et correctifs

Connection pool exhausted

Tune pool size in config.json; kill zombie sessions on DB side; verify no runaway agent loops

Vérifier : SELECT * FROM pg_stat_activity (for Postgres)

Oracle wallet auth fails

TNS_ADMIN path must be readable by the MCP process; on mac/linux watch for SELinux/AppArmor

Tool names don't appear for one DB

That connection likely failed to init; check server logs — usually wrong creds or firewall

Alternatives

db-mcp-server vs autres

Alternative	Quand l'utiliser	Compromis
postgres-mcp (official)	You only need Postgres	Single-DB
mysql-mcp (community)	You only need MySQL	Single-DB

Plus

Ressources

📖 Lire le README officiel sur GitHub

🐙 Voir les issues ouvertes

🔍 Parcourir les 400+ serveurs MCP et Skills