What is OE MCP Server?

OE MCP Server is a standalone binary that implements the Model Context Protocol (MCP) and exposes your enterprise data sources as tools that AI apps can use directly.

Connect Claude Code, Cursor, Windsurf, Codex, Claude Desktop, or VS Code to your PostgreSQL database, local filesystem, GitHub, Slack, Google Drive, SSH servers, and more — without writing any integration code.

• No code. Define connectors in a single JSON file.
• 45+ connector categories. Enterprise systems supported out of the box.
• Two transport modes. --stdio for Claude Code, Cursor, Windsurf, Codex, and Claude Desktop (launched as a child process); --serve for cloud deployments or sharing one server across a team.
• Persistent memory. Built-in memory_set / memory_get / memory_list / memory_delete tools — context survives across sessions.
• Self-hosted. Runs on your own machine. No cloud dependency. Own your data.

Setup in 3 Steps

1. Create oe-mcp.json — define your connectors (databases, files, APIs, and more).
2. Register OE MCP — add to your AI app's MCP config using --stdio (Claude Code, Cursor, Windsurf, Codex, Claude Desktop, VS Code), or start with --serve for cloud or team deployments.
3. Test — ask Claude "What connectors do you have access to?" and try saving a memory.

Quick Start via npm (Recommended)

No binary download needed — npx handles everything automatically.

Add to your AI app's MCP config (Claude Code, Cursor, Windsurf, Codex, Claude Desktop, VS Code)

macOS / Linux:

{
  "mcpServers": {
    "oe-mcp": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@openenthrium/oe-mcp", "--stdio", "/path/to/oe-mcp.json"]
    }
  }
}

Windows:

{
  "mcpServers": {
    "oe-mcp": {
      "type": "stdio",
      "command": "npx.cmd",
      "args": ["-y", "@openenthrium/oe-mcp", "--stdio", "C:\\path\\to\\oe-mcp.json"]
    }
  }
}

Note: -y tells npx to skip the install confirmation prompt. Without it, npx waits for keyboard input and the MCP connection never opens.

Reload your AI app — done.

Download (Standalone Binary)

Prefer a standalone binary? Download for your platform:

Quick Start (Binary)

1. Download the binary for your OS

# Linux / macOS — make executable
chmod +x oe-mcp-linux

2. Create your config file (oe-mcp.json)

{
  "connectors": [
    {
      "name": "my-postgres",
      "type": "postgresql",
      "host": "localhost",
      "port": 5432,
      "database": "mydb",
      "user": "postgres",
      "password": "secret"
    },
    {
      "name": "my-codebase",
      "type": "filesystem",
      "basePath": "/home/user/projects/myapp"
    }
  ],
  "memory": [
    { "key": "project_context", "value": "This is our main application database." }
  ]
}

3. Add to your AI app's MCP config (Claude Code, Cursor, Windsurf, Codex, Claude Desktop, VS Code)

{
  "mcpServers": {
    "oe-mcp": {
      "type": "stdio",
      "command": "/path/to/oe-mcp-win.exe",
      "args": ["--stdio", "/path/to/oe-mcp.json"]
    }
  }
}

Reload your AI app — the MCP tools appear automatically.

Test Your Connection and Memory

Test Connectors

Once connected, ask Claude in plain language:

"What connectors do you have access to?"

Claude will list every connected tool with its available actions. Example response:

You can also run /mcp in Claude Code to see the server status and total tool count.

Test Memory

OE MCP has built-in persistent memory that survives restarts. Use plain language or direct tool calls:

Save a memory:

"Remember that our production database host is prod-db.company.com"

Claude calls memory_set with key = main_db_host, value = prod-db.company.com.

Retrieve a memory:

"What is our production database host?"

Claude calls memory_get with key = main_db_host and returns the stored value.

List all memories:

"What do you remember about our project?"

Claude calls memory_list and returns all stored key-value pairs.

Delete a memory:

"Forget the production database host."

Claude calls memory_delete with key = main_db_host to remove it.

Memory is stored in oe-mcp-memory.json next to your oe-mcp.json and persists across sessions and restarts.

HTTP Mode (Cloud / Team Deployments)

Use --serve when you want to run OE MCP as a standalone HTTP server — for cloud deployments or sharing one server across a team.

# Start the MCP server
oe-mcp-win.exe --serve --port 4040 oe-mcp.json
# OE MCP Server listening on http://localhost:4040/mcp

In Cursor settings → MCP → Add server:

http://localhost:4040/mcp

In Claude Desktop claude_desktop_config.json:

{
  "mcpServers": {
    "oe-mcp": {
      "url": "http://localhost:4040/mcp"
    }
  }
}

Cloud Deployment (MCP as a Service)

Deploy oe-mcp-linux to any cloud server — AWS EC2, fly.io, Railway, DigitalOcean — and multiple developers connect to it via URL. No binary needed on each developer machine.

# On your cloud server
./oe-mcp-linux --serve --port 4040 /etc/oe-mcp/oe-mcp.json

Each developer adds to their Cursor / Windsurf:

http://your-server.com:4040/mcp

Config File Reference (oe-mcp.json)

{
  "connectors": [
    {
      "name": "<display-name>",
      "type": "<connection-type>",
      "...": "connector-specific credentials"
    }
  ],
  "memory": [
    { "key": "<key>", "value": "<value>" }
  ]
}

Example — Multiple Connectors

{
  "connectors": [
    { "name": "my-postgres",  "type": "postgresql",     "host": "db.company.com", "port": 5432, "database": "production", "user": "readonly", "password": "secret" },
    { "name": "my-mysql",     "type": "mysql",          "host": "localhost",       "port": 3306, "database": "mydb",       "user": "root",     "password": "secret" },
    { "name": "my-mongo",     "type": "mongodb",        "uri": "mongodb://localhost:27017",       "database": "mydb" },
    { "name": "my-redis",     "type": "redis",          "host": "localhost",       "port": 6379 },
    { "name": "my-elastic",   "type": "elasticsearch",  "node": "https://localhost:9200",         "apiKey": "xxxxxxxxxxxx" },
    { "name": "my-s3",        "type": "s3",             "accessKeyId": "AKIAXXXXXXXX",            "secretAccessKey": "xxxxxxxxxxxx", "region": "us-east-1", "bucket": "my-bucket" },
    { "name": "my-gdrive",    "type": "gdrive",         "clientId": "xxxx.apps.googleusercontent.com", "clientSecret": "xxxx", "refreshToken": "xxxx" },
    { "name": "my-github",    "type": "github",         "repoUrl": "https://github.com/your-org/your-repo", "personalAccessToken": "ghp_xxxxxxxxxxxx" },
    { "name": "my-jira",      "type": "jira",           "host": "https://company.atlassian.net",  "email": "you@company.com", "apiToken": "xxxx" },
    { "name": "my-slack",     "type": "slack",          "botToken": "xoxb-xxxxxxxxxxxx" },
    { "name": "my-gmail",     "type": "gmail",          "clientId": "xxxx.apps.googleusercontent.com", "clientSecret": "xxxx", "refreshToken": "xxxx" },
    { "name": "my-smtp",      "type": "smtp",           "host": "smtp.company.com", "port": 587,  "user": "you@company.com", "password": "secret" },
    { "name": "my-server",    "type": "ssh",            "host": "server.company.com", "port": 22, "username": "ubuntu", "privateKey": "-----BEGIN OPENSSH PRIVATE KEY-----\nYOUR_PRIVATE_KEY_CONTENT\n-----END OPENSSH PRIVATE KEY-----" },
    { "name": "my-codebase",  "type": "filesystem",    "basePath": "/home/user/projects" },
    { "name": "my-api",       "type": "rest-api",       "baseUrl": "https://api.company.com",     "headers": { "Authorization": "Bearer xxxx" } },
    { "name": "my-hubspot",   "type": "hubspot",        "accessToken": "pat-xxxxxxxxxxxx" },
    { "name": "my-kafka",     "type": "kafka",          "brokers": ["localhost:9092"] }
  ],
  "memory": [
    { "key": "team",        "value": "Platform Engineering" },
    { "key": "environment", "value": "production" }
  ]
}

Built-in Tools

Connector Tools

Each connector exposes a set of tools prefixed with the connector name. Examples:

Memory Tools

Built-in memory tools available in every session:

Memory is stored in oe-mcp-memory.json next to your oe-mcp.json and survives restarts.

Example usage:

"Remember that our main database is on prod-db.company.com" → Claude calls memory_set with key main_db_host and value prod-db.company.com

Action Log Tools

Built-in log tools that record every connector tool call:

Every connector tool call is automatically appended to oe-mcp-log.json next to your oe-mcp.json with timestamp, connector name, tool, input, and result. Memory and log tool calls are excluded.

Example usage:

"Show me the action log" → Claude calls log_list and returns recent connector activity

Example log entry:

{
  "ts": "2026-08-08T04:59:33.289Z",
  "connector": "my-postgres",
  "tool": "query",
  "input": { "sql": "SELECT * FROM users LIMIT 10" },
  "result": "ok"
}

Agent Runner Tool

OE MCP can run OE Runtime YAML agents directly from Claude Code, Cursor, Windsurf, Codex, or any MCP-compatible AI app — no terminal required.

Parameters:

Config auto-detection: OE MCP looks for oe-config.json in the same directory as agent.yaml. If found, it uses that config (correct LLM + connector credentials for that agent). Otherwise it falls back to the oe-mcp.json config.

Example — ask Claude to run an agent:

"Run my database analyst agent at /home/user/agents/db-analyst/agent.yaml" → Claude calls run_agent with file = /home/user/agents/db-analyst/agent.yaml

Example — run with params:

"Run my report agent for the month of July" → Claude calls run_agent with file = /agents/report.yaml and params = { "month": "July" }

What happens under the hood:

npx -y @openenthrium/oe-runtime agent.yaml --config oe-config.json [--param key=value ...]

OE Runtime executes the YAML agent — calling connectors, running LLM steps, and returning the full output back to your AI app.

Requires OE Runtime. The agent directory must have a valid oe-config.json with llm and connectors configured. See OE Runtime for agent authoring docs.

Binary vs Node.js Mode

The standalone binary works for all connector categories except Oracle, MSSQL, SQLite, and Snowflake — these use native C++ addons that cannot be bundled into a single executable.

If you need any of these four, run with Node.js instead:

git clone https://github.com/enthrium/open-enthrium-ai-mcp-server.git
cd open-enthrium-ai-mcp-server/server
yarn install
# stdio mode (Claude Code, Cursor, Windsurf, Codex, Claude Desktop, VS Code)
node mcp/index.js --stdio /path/to/oe-mcp.json
# serve mode (cloud/team deployments)
node mcp/index.js --serve --port 4040 /path/to/oe-mcp.json

All other connectors (PostgreSQL, MySQL, MongoDB, Redis, S3, Slack, GitHub, REST API, SSH, filesystem, etc.) work directly with the binary — no Node.js required.

Connector Catalog

Connectors across 45+ categories:

Sample Configs

Download oe-mcp-samples.zip for ready-to-use configs:

postgres · mysql · mongodb · github · slack · gdrive · ssh · filesystem · oracle · salesforce · servicenow · telegram · notion · confluence · graphql · zoho-mail · sftp · dropbox · multi-connector

Each sample includes the complete oe-mcp.json with setup instructions in comments.

Transport Modes

Both modes are supported in the same binary — just pass the appropriate flag.

Part of Open Enthrium

OE MCP Server is part of the Open Enthrium platform.

License

Apache-2.0 — free to use, modify, and deploy for any purpose, including commercial use. No usage limits. No telemetry. No call-home.