
no2sql
io.github.Ptalati015/no2sql
No2SQL MCP server for inferring SQL schema and ER diagrams from MongoDB databases.
Documentation
No2SQL
No2SQL is an MCP server that helps you move from MongoDB to SQL. It lets your AI assistant:
- discover collections and likely relationships,
- generate SQL schema scripts,
- generate SQL seed (INSERT) statements,
- draw ER diagrams in Mermaid, PlantUML and GraphViz DOT.
It works with any AI client that runs local MCP servers, including Claude Desktop, Claude Code, VS Code, Cursor, Windsurf, OpenAI Codex CLI and the OpenAI Agents SDK.
Who This Is For
- Teams migrating from MongoDB to relational databases
- Engineers documenting NoSQL structures in ERD form
- AI-assisted workflows where an assistant calls data-migration tools through MCP
Install
No2SQL is a local MCP server. Your AI client starts it on your own machine and talks to it over stdio. There are two ways to install it.
Option A: download the zip (no .NET needed)
-
Download the zip for your platform from the v1.1.1 release:
Platform File Windows x64 No2SQL-1.1.1-win-x64.zipWindows ARM64 No2SQL-1.1.1-win-arm64.zipmacOS Apple silicon No2SQL-1.1.1-osx-arm64.zipmacOS Intel No2SQL-1.1.1-osx-x64.zipLinux x64 No2SQL-1.1.1-linux-x64.zipLinux ARM64 No2SQL-1.1.1-linux-arm64.zipAlso download
SHA256SUMS.txtfrom the same release. -
Verify the download (recommended). Run this in the folder that holds both files:
# Linux sha256sum -c SHA256SUMS.txt --ignore-missing # macOS shasum -a 256 -c SHA256SUMS.txt --ignore-missing# Windows: compare the output with the matching line in SHA256SUMS.txt (Get-FileHash .\No2SQL-1.1.1-win-x64.zip -Algorithm SHA256).HashEvery release also has a build provenance attestation. If you have the GitHub CLI, you can check that the zip was built by this repository's release workflow. Sign in with
gh auth loginfirst:gh attestation verify No2SQL-1.1.1-<platform>.zip --repo Ptalati015/No2SQL -
macOS only: the exe is not notarized, so Gatekeeper blocks it the first time. Clear the quarantine flag on the one file you downloaded, before you unzip it:
xattr -d com.apple.quarantine No2SQL-1.1.1-osx-arm64.zipIf you already unzipped it, run the same command on the exe instead (
xattr -d com.apple.quarantine /path/to/No2SQL). Don't turn Gatekeeper off. -
Unzip it somewhere permanent, such as
~/tools/no2sqlorC:\tools\no2sql. The zip holds one file to run,No2SQL(No2SQL.exeon Windows), next toLICENSE,NOTICEandTRADEMARK.md. On Linux and macOS it is already executable.
Your client config uses the absolute path to that file as its command, with no args.
Option B: run it with dnx (.NET 10 SDK)
If you have the .NET 10 SDK, dnx downloads No2SQL from NuGet and runs it, so there is nothing to unzip:
dnx No2SQL@1.1.1 --yes
You don't run this yourself. Your client config uses dnx as its command and ["No2SQL@1.1.1", "--yes"] as its args. If a client says it can't find dnx (Claude Desktop on macOS often starts with a short PATH), use the full path from which dnx (or where dnx on Windows).
Configure
No2SQL reads these environment variables. Set them in your MCP client's env config, as the examples below show. Never pass the connection string as a command-line argument.
| Variable | Required | What it does |
|---|---|---|
NO2SQL_MONGO | Yes | MongoDB connection string, for example mongodb+srv://<user>:<password>@<cluster-host>/. It contains credentials, so treat it as a secret. |
NO2SQL_ALLOWED_DATABASES | Recommended | Comma-separated list of the databases No2SQL may read, for example sales,inventory. If it is empty, every database is readable except the system ones. |
NO2SQL_BLOCK_SYSTEM_DATABASES | No | Defaults to true, which blocks admin, config and local. Leave it on. |
Without NO2SQL_MONGO the server starts and lists its tools, but every tool call, including test_connectivity, returns an error.
Set up your AI client
Each example uses the zip install on macOS or Linux (/Users/you/tools/no2sql/No2SQL). Change the path to where you unzipped it. On Windows, escape the backslashes in JSON ("C:\\tools\\no2sql\\No2SQL.exe").
Using dnx instead? Replace the command and args lines in any example with:
"command": "dnx",
"args": ["No2SQL@1.1.1", "--yes"],
Claude Desktop
Open Settings → Developer → Edit Config, or edit the file directly:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"no2sql": {
"command": "/Users/you/tools/no2sql/No2SQL",
"args": [],
"env": {
"NO2SQL_MONGO": "mongodb+srv://<user>:<password>@<cluster-host>/",
"NO2SQL_ALLOWED_DATABASES": "mydb"
}
}
}
}
Quit and reopen Claude Desktop. No2SQL's tools then appear in the tools menu of the chat box.
Claude Code
Add a .mcp.json file to your project root. Claude Code expands ${NO2SQL_MONGO} from your shell environment, so the file holds no secret and is safe to commit:
{
"mcpServers": {
"no2sql": {
"command": "/Users/you/tools/no2sql/No2SQL",
"args": [],
"env": {
"NO2SQL_MONGO": "${NO2SQL_MONGO}",
"NO2SQL_ALLOWED_DATABASES": "mydb"
}
}
}
}
Set NO2SQL_MONGO in your shell profile (for example ~/.zshrc), then start claude in that project and approve the server when asked. Check it with claude mcp list, or with /mcp inside a session. Never put the real connection string in a .mcp.json that you commit.
VS Code (GitHub Copilot agent mode)
Needs VS Code 1.102 or later. Create .vscode/mcp.json in your workspace, or run MCP: Open User Configuration for all workspaces. VS Code asks for the connection string the first time the server starts and stores it securely, so it never sits in the file:
{
"inputs": [
{
"type": "promptString",
"id": "no2sql-mongo",
"description": "MongoDB connection string for No2SQL",
"password": true
}
],
"servers": {
"no2sql": {
"type": "stdio",
"command": "/Users/you/tools/no2sql/No2SQL",
"args": [],
"env": {
"NO2SQL_MONGO": "${input:no2sql-mongo}",
"NO2SQL_ALLOWED_DATABASES": "mydb"
}
}
}
}
Run MCP: List Servers to start it, then open Copilot Chat in Agent mode and check that No2SQL's tools are enabled under the tools icon. Older guides show an "mcp.servers" block in settings.json; use mcp.json as shown here instead.
Cursor
Edit ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project):
{
"mcpServers": {
"no2sql": {
"command": "/Users/you/tools/no2sql/No2SQL",
"args": [],
"env": {
"NO2SQL_MONGO": "mongodb+srv://<user>:<password>@<cluster-host>/",
"NO2SQL_ALLOWED_DATABASES": "mydb"
}
}
}
}
Open Cursor Settings → MCP and check that no2sql shows a green dot. Don't commit a project .cursor/mcp.json that holds a real connection string.
Windsurf
Edit ~/.codeium/windsurf/mcp_config.json, or open it from the MCP servers panel in Cascade:
{
"mcpServers": {
"no2sql": {
"command": "/Users/you/tools/no2sql/No2SQL",
"args": [],
"env": {
"NO2SQL_MONGO": "mongodb+srv://<user>:<password>@<cluster-host>/",
"NO2SQL_ALLOWED_DATABASES": "mydb"
}
}
}
}
Press the refresh button in the MCP panel to load it.
OpenAI Codex CLI
Add this to ~/.codex/config.toml:
[mcp_servers.no2sql]
command = "/Users/you/tools/no2sql/No2SQL"
args = []
env = { NO2SQL_MONGO = "mongodb+srv://<user>:<password>@<cluster-host>/", NO2SQL_ALLOWED_DATABASES = "mydb" }
For dnx, use command = "dnx" and args = ["No2SQL@1.1.1", "--yes"]. Start codex and run /mcp to see the server.
OpenAI Agents SDK (Python)
The Agents SDK can start local stdio servers itself. This example reads the connection string from your environment:
import asyncio
import os
from agents import Agent, Runner
from agents.mcp import MCPServerStdio
async def main():
async with MCPServerStdio(
name="No2SQL",
params={
"command": "/Users/you/tools/no2sql/No2SQL",
"args": [],
"env": {
"NO2SQL_MONGO": os.environ["NO2SQL_MONGO"],
"NO2SQL_ALLOWED_DATABASES": "mydb",
},
},
) as no2sql:
agent = Agent(
name="Migration helper",
instructions="Use the No2SQL tools to analyze MongoDB and produce SQL.",
mcp_servers=[no2sql],
)
result = await Runner.run(agent, "Generate a SQL schema for the mydb database.")
print(result.final_output)
asyncio.run(main())
Not supported yet
ChatGPT, claude.ai on the web and mobile, and the hosted Claude and OpenAI APIs (the Responses API mcp tool and the Claude API MCP connector) can only reach remote MCP servers over HTTPS. No2SQL runs locally over stdio for now. Remote mode is planned for a later release.
Try it
Once your client shows the No2SQL tools, ask:
Use No2SQL to list my MongoDB databases.
Infer the relationships in database mydb.
Generate a SQL schema for mydb.
Generate a Mermaid ERD for mydb.
Generate SQL INSERT statements for the orders collection in mydb.
MCP tools
All tools only read from MongoDB. Each one carries MCP tool annotations that mark it read-only, non-destructive and idempotent, so clients can tell it is safe to run.
| Tool | What it does |
|---|---|
test_connectivity | Checks that the server is running. It doesn't contact MongoDB, but like every tool it needs NO2SQL_MONGO to be set. |
list_databases | Lists the databases you can analyze, filtered by NO2SQL_ALLOWED_DATABASES. |
list_inferred_relationships | Lists the ID-like fields (such as userId) in each collection. |
compare_id_fields_to_ids | Infers relationships by matching ID-like field values to _id values, with a confidence score. |
generate_sql_schema | Generates a SQL script with CREATE TABLE statements, indexes and foreign keys. |
generate_sql_schema_advanced | Same as above, plus relationships you supply yourself, returned as structured output. |
generate_seeders_for_collection | Generates SQL INSERT statements for up to 1000 documents of one collection. |
generate_erd_mermaid | Generates a Mermaid ER diagram (source: sql, auto or mongo). |
generate_erd_plant_uml | Generates a PlantUML ER diagram (source: sql, auto or mongo). |
generate_erd_dot | Generates a GraphViz DOT ER diagram (source: sql, auto or mongo). |
Security guardrails
No2SQL validates every tool input: identifier format and length, source values, relationship override fields and prompt-injection marker patterns. Errors sent to the client never include connection details. When a tool fails while running, the client gets an error id and the details go to the server log.
Recommended baseline:
- Set
NO2SQL_ALLOWED_DATABASESto only the databases you want analyzed. - Keep
NO2SQL_BLOCK_SYSTEM_DATABASES=true. - Use a MongoDB user with read-only permissions.
- Keep the connection string in your client's
envconfig or a secret prompt, never on a command line or in a committed file.
Build, Test, Publish
Build solution:
dotnet build No2SQL.sln
Run test harness:
dotnet run --project No2SQL.Test
Pack NuGet:
dotnet pack -c Release
This produces the primary No2SQL tool package plus one self-contained package per platform, which is what dnx installs.
Publish NuGet package:
dotnet nuget push **/bin/Release/*.nupkg --api-key <your-api-key> --source https://api.nuget.org/v3/index.json
Run from source: run dotnet run from the No2SQL folder. When you run from source you can also set the connection string under ConnectionStrings:MongoDb in No2SQL/appsettings.json, but don't commit real values. That file is only read when the working directory is No2SQL, so dotnet run --project No2SQL from the repo root ignores it.
Local Test Defaults
No2SQL.Test/Program.cs supports environment variable overrides:
| Variable | Description |
|---|---|
NO2SQL_TEST_DATABASE | Database used when none is supplied |
NO2SQL_TEST_COLLECTION | Collection used by --test-stream when none is supplied |
NO2SQL_TEST_ROWS_PER_CHUNK | Optional chunk-size override for stream seeder checks |
NO2SQL_TEST_LIMIT | Optional max document count override for stream seeder checks |
NO2SQL_TEST_BATCH_SIZE | Optional Mongo cursor batch size override for stream seeder checks |
CLI arguments take precedence over environment variables.
Contributions
Forks and derivative works are welcome under the MIT License.
Changes to the official No2SQL repository are accepted at the maintainer's discretion. Submitting a pull request does not guarantee inclusion.
License
No2SQL is licensed under the MIT License.
Copyright (c) 2026 Preet Talati
"No2SQL" is a trademark of Preet Talati.
No2SQL# nuget: No2SQL