MongoDB MCP Server

Name: MongoDB MCP Server MCP Server
Author: kiliczsh

Why use it

Key features

Pass connection string as CLI arg — no env-file dance
Read-only mode disables all mutating ops
Schema inference from sampled docs
Aggregation pipeline support
Index inspection for query-tuning conversations

Live Demo

What it looks like in practice

mongo-mcp-kiliczsh.replay ▶ ready

0/0

Install

Pick your client

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

{
  "mcpServers": {
    "mongo-mcp-kiliczsh": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-mongo-server",
        "mongodb://localhost:27017/mydb"
      ]
    }
  }
}

Open Claude Desktop → Settings → Developer → Edit Config. Restart after saving.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "mongo-mcp-kiliczsh": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-mongo-server",
        "mongodb://localhost:27017/mydb"
      ]
    }
  }
}

Cursor uses the same mcpServers schema as Claude Desktop. Project config wins over global.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "mongo-mcp-kiliczsh": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-mongo-server",
        "mongodb://localhost:27017/mydb"
      ]
    }
  }
}

Click the MCP Servers icon in the Cline sidebar, then "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "mongo-mcp-kiliczsh": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-mongo-server",
        "mongodb://localhost:27017/mydb"
      ]
    }
  }
}

Same shape as Claude Desktop. Restart Windsurf to pick up changes.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "mongo-mcp-kiliczsh",
      "command": "npx",
      "args": [
        "-y",
        "mcp-mongo-server",
        "mongodb://localhost:27017/mydb"
      ]
    }
  ]
}

Continue uses an array of server objects rather than a map.

~/.config/zed/settings.json

{
  "context_servers": {
    "mongo-mcp-kiliczsh": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "mcp-mongo-server",
          "mongodb://localhost:27017/mydb"
        ]
      }
    }
  }
}

Add to context_servers. Zed hot-reloads on save.

claude mcp add mongo-mcp-kiliczsh -- npx -y mcp-mongo-server mongodb://localhost:27017/mydb

One-liner. Verify with claude mcp list. Remove with claude mcp remove.

Use Cases

Real-world ways to use MongoDB MCP Server

Explore an unfamiliar MongoDB database without writing anything

👤 Devs onboarding to a new service ⏱ ~15 min beginner

When to use: You inherited a Mongo instance and the schema is undocumented.

Prerequisites

Read-only Mongo user — db.createUser({user:'reader', roles:[{role:'read', db:'mydb'}]})

Flow

Connect read-only

Use the mongo MCP. List all collections and infer the schema for the top 3 by size.✓ Copied

→ Collection list + JSON schema per collection
Spot-check

Show me 5 sample documents from orders — anonymize emails.✓ Copied

→ 5 docs, PII redacted
Map relationships

Which collections reference each other by ObjectId? Draw a quick text diagram.✓ Copied

→ Plain-text ER diagram

Outcome: Working mental model of the database in 10 minutes, no DBA needed.

Pitfalls

Sampling miss high-cardinality fields — Increase sample size in inference call; sample by date if data is time-skewed

Combine with: filesystem

Tune a slow aggregation with index advice from Claude

👤 Backend devs ⏱ ~25 min intermediate

When to use: An aggregation pipeline takes 30s and you don't know which stage is the cost.

Flow

Profile

Run this aggregation with explain=true. Show the winning plan.✓ Copied

→ Explain output with stage docs scanned
Diagnose

Which stage is the bottleneck? What index would help?✓ Copied

→ Concrete index suggestion with rationale
Verify

Re-explain after I add the index. Did totalDocsExamined drop?✓ Copied

→ Yes/no with numbers

Outcome: Aggregation that's 10x faster with one targeted index.

Pitfalls

Claude suggests an index that bloats writes — Ask: 'What write penalty does this index add?' before applying

Find documents that violate an implicit schema

👤 Data engineers ⏱ ~20 min intermediate

When to use: Schemaless DBs accumulate drift; you need to find offenders.

Flow

Sample

Sample 1000 documents from users. What fields are missing or have unexpected types?✓ Copied

→ Per-field nullity / type-frequency table
Find

Find all users where email is null or not a string.✓ Copied

→ Count + sample _ids

Outcome: Concrete dirty-data report you can hand to migration scripts.

Pitfalls

Large collections hit timeouts — Use $sample then scope follow-ups by ranges

Combinations

Pair with other MCPs for X10 leverage

mongo-mcp-kiliczsh + filesystem

Export aggregation results to JSON for downstream tools

Run the aggregation and save results to /tmp/orders-by-month.json.✓ Copied

mongo-mcp-kiliczsh + github

Open a PR with a migration script after spotting drift

We found 1200 users without email. Open a PR with a backfill migration.✓ Copied

Tools

What this MCP exposes

Tool	Inputs	When to call	Cost
find	collection, filter, projection?, limit?	Look up specific docs	1 query
aggregate	collection, pipeline[], explain?	Group/transform data	1 query
list_collections	(none)	Discover what's in the DB	free
schema	collection, sample_size?	Infer shape from sampled docs	1 sample read
list_indexes	collection	Tuning conversations	free
insert_one	collection, document	Only when not in --read-only mode	1 write

Cost & Limits

What this costs to run

API quota: Bound by your Mongo cluster's RU/IOPS
Tokens per call: 200–5000 depending on result size
Monetary: Free (open source)
Tip: Project only the fields you need; cap limit early

Security

Permissions, secrets, blast radius

Minimum scopes: read on the target db

Credential storage: URI passed as CLI arg — keep it out of shell history

Data egress: Only to your Mongo host

Never grant: dbAdmin root

URIs with passwords end up in process args; some systems log those

Troubleshooting

Common errors and fixes

MongoServerSelectionError

Check connectivity, IP allowlist, TLS settings

Verify: mongosh with the same URI

Authentication failed

Check authSource in URI (often admin); confirm user exists in that DB

Tool not found in read-only mode

Drop --read-only if you need writes; otherwise use a different MCP for writes

Alternatives