OE MCP Server · @openenthrium/oe-mcp

Connect Claude Code, Cursor, Windsurf, Codex, Claude Desktop, and VS Code to enterprise data sources — databases, APIs, files, SSH, messaging, and more. One binary. One JSON config.

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.
• Action log. Built-in log_list / log_clear tools — every connector call is automatically recorded with timestamp, connector, tool, input, and result.
• Run AI agents. run_agent executes any OE Runtime YAML agent directly from Claude Code, Cursor, Windsurf, or any MCP-enabled AI chat — no terminal required.
• Agent chains. Chain agents together in YAML — auto chains fire in sequence and return nested results; manual chains pause for human approval via approve_chain; works in Claude Code, Cursor, Telegram, or any MCP client.
• Self-hosted. Runs on your own machine. No cloud dependency. Own your data.

Quick Start

1. 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-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-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-slack",    "type": "slack",         "botToken": "xoxb-xxxxxxxxxxxx" },
    { "name": "my-gmail",    "type": "gmail",         "clientId": "xxxx.apps.googleusercontent.com", "clientSecret": "xxxx", "refreshToken": "xxxx" },
    { "name": "my-email",    "type": "smtp",          "host": "smtp.company.com", "port": 587, "user": "[email protected]", "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/myapp" },
    { "name": "my-api",      "type": "rest-api",      "baseUrl": "https://api.company.com", "headers": { "Authorization": "Bearer xxxx" } },
    { "name": "my-jira",     "type": "jira",          "host": "https://company.atlassian.net", "email": "[email protected]", "apiToken": "xxxx" },
    { "name": "my-hubspot",  "type": "hubspot",       "accessToken": "pat-xxxxxxxxxxxx" },
    { "name": "my-kafka",    "type": "kafka",         "brokers": ["localhost:9092"] },
    { "name": "my-elastic",  "type": "elasticsearch", "node": "https://localhost:9200", "apiKey": "xxxxxxxxxxxx" }
  ],
  "memory": [
    { "key": "team",        "value": "Platform Engineering" },
    { "key": "environment", "value": "production" }
  ]
}

2. 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 skips npx's install confirmation prompt — without it, npx blocks waiting for keyboard input and the MCP connection never opens.

3. Reload your AI app — your connectors appear as tools 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:

| Connector | Tools | |---|---| | my-postgres | query | | my-github | list_files, read_file, create_issue, get_issue, search_issues | | my-slack | list_channels, post_message, search_messages | | my-codebase | list_dir, read_file, write_file, search_files |

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.

Run YAML Agents

OE MCP can run OE Runtime YAML agents directly from Claude Code, Cursor, Windsurf, Codex, or any MCP-compatible AI app — no terminal required. Agents can chain to other agents, with auto or manual approval.

| Tool | Description | |---|---| | run_agent | Run a YAML agent. Returns output, auto-chain results, and any pending manual chains. | | list_pending_chains | List all manual chains waiting for approval — chain_id, next agent, and output preview. | | approve_chain | Approve or reject a pending manual chain by chain_id. Runs immediately and returns full output. |

Just ask Claude:

"Run my security monitor at /agents/security-monitor.yaml" → Claude calls run_agent → output returned, pending manual chains listed

"Approve the chain" → Claude calls approve_chain → chained agent runs → output returned

run_agent parameters:

| Parameter | Type | Required | Description | |---|---|---|---| | file | string | ✅ | Absolute path to the agent.yaml file | | params | object | ❌ | Key-value pairs substituted into the agent prompt via {{key}} | | input | string | ❌ | Optional initial message or context passed to the agent |

Agent chain YAML syntax:

chains:
  - next_agent: ./followup.yaml     # relative path from this agent file
    trigger_type: auto              # fires immediately after this agent completes

  - next_agent: ./notify.yaml
    trigger_type: manual            # pauses — Claude asks you before running

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

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

Supported Connectors

Connectors across 45+ categories:

| Category | Examples | |---|---| | SQL Databases | PostgreSQL, MySQL, MSSQL, Oracle, SQLite, Snowflake, BigQuery, Redshift | | NoSQL / Cache | MongoDB, Redis, Elasticsearch, DynamoDB, Cassandra | | Object Storage | AWS S3, GCS, Azure Blob, MinIO, Cloudflare R2 | | Cloud Drives | Google Drive, OneDrive, Dropbox, Box | | Filesystem | Local directories — list, read, write, search | | Email | Gmail, Outlook, Zoho Mail, SMTP | | Team Messaging | Slack, Microsoft Teams, Discord, Telegram | | CRM | HubSpot, Salesforce, Notion, Airtable | | Issue Tracking | GitHub, Jira, GitLab, Linear | | REST API | Any HTTP/REST endpoint | | SSH / SFTP | Remote command execution, file transfer | | Message Queues | Kafka, AWS SQS, Google Pub/Sub, RabbitMQ | | + more | LDAP, OCR, Image Generation, Healthcare, ERP, Web3, ... |

Transport Modes

| Mode | Flag | Best for | |---|---|---| | stdio | --stdio | Claude Code, Cursor, Windsurf, Codex, Claude Desktop — binary launched as child process by the AI app | | HTTP | --serve --port 4040 | Cloud deployments, multiple developers sharing one server |

Built-in Memory

Every session includes persistent memory tools:

| Tool | Description | |---|---| | memory_set | Store a key-value pair across sessions | | memory_get | Retrieve a stored value | | memory_list | List all stored pairs | | memory_delete | Remove a stored key |

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

Built-in Action Log

Every connector tool call is automatically logged to oe-mcp-log.json:

| Tool | Description | |---|---| | log_list | List recent connector action log entries (newest first, supports limit param) | | log_clear | Clear all entries from the action log |

Example log entry:

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

Links

• Full Documentation
• GitHub Repository
• Discord Community
• OE Platform — full web app with Agent Builder, RAG, workspaces

License

Apache-2.0 — free to use, modify, and deploy for any purpose including commercial use.