# Db2 for i MCP Server
[](https://github.com/Strom-Capital/mcp-server-db2i/actions/workflows/ci.yml)
[](https://www.npmjs.com/package/mcp-server-db2i)
[](https://opensource.org/licenses/MIT)
[](https://modelcontextprotocol.io/)
[](https://www.ibm.com/products/ibm-i)
[](https://www.typescriptlang.org/)
[](https://nodejs.org/)
[](docs/docker.md)
[](https://www.npmjs.com/package/mcp-server-db2i)
[](https://github.com/sponsors/Strom-Capital)
`mcp-server-db2i` is a [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) server for IBM Db2 for i (Db2i) on IBM i (AS/400). It enables AI assistants like Claude and Cursor to query and inspect IBM i databases through the IBM i Access ODBC driver, or optionally the JT400 JDBC driver or Mapepire over SSH.
Listed in the [MCP Registry](https://registry.modelcontextprotocol.io/) as `io.github.Strom-Capital/mcp-server-db2i`.
**Website:** [db2i-mcp.com](https://db2i-mcp.com), with a [blog](https://db2i-mcp.com/blog) that explains each release. **Docs:** [docs.db2i-mcp.com](https://docs.db2i-mcp.com).
## Architecture
AI clients connect to the MCP Server in one of two ways. Local clients such as Claude Desktop, Claude Code and Cursor can start it as a process and talk over stdio. Remote clients connect over Streamable HTTP at `/mcp`, signing in with OAuth 2.1 (claude.ai custom connectors) or a bearer token (custom agents). Local clients can also use the HTTP endpoint. The server executes read-only queries against Db2 for i using the IBM i Access ODBC driver (default, no Java), the optional JT400 JDBC driver (`DB2I_DRIVER=jt400`), or Mapepire over SSH (`DB2I_DRIVER=mapepire`) for systems where only SSH is reachable. One server can reach several IBM i systems through connection profiles, each with its own driver.
```mermaid
graph LR
subgraph clients ["AI Clients"]
local("Claude Desktop, Claude Code, Cursor")
remote("claude.ai connectors")
agents("Custom Agents")
end
subgraph server ["MCP Server"]
stdio["stdio"]
http["Streamable HTTP + Auth"]
tools[["MCP Tools"]]
profiles{{"System profiles"}}
odbc["IBM i Access ODBC"]
jdbc["JT400 JDBC (optional)"]
mapepire["Mapepire over SSH (optional)"]
end
subgraph prod ["IBM i: prod"]
db2prod[("Db2 for i")]
end
subgraph test ["IBM i: test"]
db2test[("Db2 for i")]
end
subgraph dev ["IBM i: dev"]
db2dev[("Db2 for i")]
end
local -->|local process| stdio
local -.->|remote URL| http
remote -->|OAuth 2.1| http
agents -->|bearer token| http
stdio & http --> tools
tools --> profiles
profiles --> odbc & jdbc & mapepire
odbc -->|ODBC| db2prod
jdbc -->|JDBC| db2test
mapepire -->|SSH| db2dev
```
## Features
- **Read-only SQL queries** - Execute SELECT statements safely with automatic result limiting, and a query timeout that cancels runaway statements on the IBM i
- **Schema inspection** - List all schemas/libraries with optional filtering
- **Table metadata** - List tables, describe columns, view indexes and constraints
- **View inspection** - List and explore database views
- **Secure by design** - Only SELECT queries allowed, credentials via environment variables
- **Docker support** - Run as a container for easy deployment
- **HTTP Transport** - MCP over Streamable HTTP with token authentication for remote clients and agents
- **OAuth for Remote Clients** - Built-in OAuth 2.1 sign-in with the user's own IBM i profile, so claude.ai custom connectors can connect. The sign-in page can carry your company's name, logo, colors and language ([Customize for your company](docs/customization.md))
- **Current MCP spec** - Speaks [2026-07-28](https://modelcontextprotocol.io/) and still serves stateless 2025-era clients
- **Dual Transport** - Run stdio and HTTP simultaneously
- **Multiple systems** - Reach several IBM i systems from one server with `DB2I_PROFILES`, each with its own driver, credentials, and library allowlist. Tools take an optional `system` argument. See [Multiple systems](docs/configuration.md#multiple-systems)
- **SSH-only systems** - With `DB2I_DRIVER=mapepire`, reach an IBM i where only SSH is open. Mapepire starts inside the SSH session, with no server install and a host key check. See [Using the Mapepire driver](docs/configuration.md#using-the-mapepire-driver-ssh)
- **Tool selection** - Enable or disable individual tools, e.g. a metadata-only mode without `execute_query`
- **Business SQL tools** - Load read-only ERP queries and table notes from YAML, and check the files with `mcp-server-db2i validate-tools` before the server starts. See [Business SQL tools](docs/custom-tools.md)
- **Server instructions** - Send the rules no query may miss, such as which flag marks a deleted row, to the model at the start of every session. See [Instructions](docs/custom-tools.md#instructions)
- **Charts in the chat** - Hosts that support MCP Apps show query results as a sortable grid under the tool call, with a chart one click away. When the user asks for a chart, or a YAML tool sets one, it opens on the chart. See [Result view](docs/tools.md#result-view)
- **Compact responses** - Compact JSON by default, or markdown tables to save tokens
- **Statement checks and DDL** - Validate object names, return the SQL that recreates an object, and list what depends on a table
- **Catalog search and profiling** - Find tables and columns across libraries, check journaling, and profile a table's row counts and value ranges
- **Column masking** - Redact sensitive columns, or show only their last four characters, in query results. See [Column masking](docs/security.md#column-masking)
- **Query exports** - Hand the user a CSV or Excel file of a query's results, as a file path over stdio or a short-lived download link over HTTP. The rows never pass through the model. See [Query exports](docs/tools.md#query-exports)
- **Audit log** - Record every tool call and sign-in as one JSON line, with the SQL hashed by default. Optionally record the model's reason for each call (`MCP_TOOL_INTENT`). See [Audit log](docs/security.md#audit-log)
- **Tool reload** - Reload YAML tool files when they change, with `MCP_CUSTOM_TOOLS_WATCH=true`
- **Resources and prompts** - Read table columns and DDL as MCP resources, and start from prompts that explore a library, explain a table, or write a query. See [Resources and prompts](#resources-and-prompts)
## Quick Start
### Installation
```bash
npm install -g mcp-server-db2i
```
The default `odbc` driver needs unixODBC and the IBM i Access ODBC Driver on the machine. No Java is needed. The other drivers' packages are not installed by default, so add them next to the server:
```bash
npm install -g mcp-server-db2i node-jt400 # DB2I_DRIVER=jt400, needs a JDK to install and a JRE to run
npm install -g mcp-server-db2i @ibm/mapepire-js ssh2 # DB2I_DRIVER=mapepire, when only SSH reaches the IBM i
```
With npx, pass them with `-p`, for example `npx -y -p mcp-server-db2i@latest -p node-jt400 mcp-server-db2i`. See [Installing the jt400 and mapepire packages](docs/configuration.md#installing-the-jt400-and-mapepire-packages).
Or with Docker:
```bash
docker build -t mcp-server-db2i . # odbc image (amd64; add --platform linux/amd64 on arm64)
docker build --target jt400 -t mcp-server-db2i . # jt400 image (builds natively on arm64)
```
### Configuration
Create a `.env` file with your IBM i credentials:
```env
DB2I_HOSTNAME=your-ibm-i-host.com
DB2I_USERNAME=your-username
DB2I_PASSWORD=your-password
DB2I_SCHEMA=your-default-schema # Optional
```
### Client Setup
Add to your MCP client config (e.g., `~/.cursor/mcp.json`):
```json
{
"mcpServers": {
"db2i": {
"command": "npx",
"args": ["-y", "mcp-server-db2i@latest"],
"env": {
"DB2I_HOSTNAME": "${env:DB2I_HOSTNAME}",
"DB2I_USERNAME": "${env:DB2I_USERNAME}",
"DB2I_PASSWORD": "${env:DB2I_PASSWORD}"
}
}
}
}
```
This uses environment variable expansion to keep credentials out of config files. Set the variables in your shell profile (`~/.zshrc` or `~/.bashrc`).
See the [Client Setup Guide](docs/client-setup.md) for Cursor, Claude Desktop, Claude Code, and Docker setup options.
No MCP client? The [example web app](examples/web-app) is a browser chat with result tables, exports and the business SQL tools. Users sign in with their IBM i user profile and bring their own model API key.
## Available Tools
| Tool | Description |
|------|-------------|
| `execute_query` | Execute read-only SELECT queries, with an optional `chart` hint for the [result view](docs/tools.md#result-view) |
| `export_query` | Write every row of a read-only query to a CSV or XLSX file: a path over stdio, a short-lived download link over HTTP. Off unless `EXPORT_ENABLED` is set |
| `list_schemas` | List schemas/libraries (with optional filter) |
| `list_tables` | List tables in a schema (with optional filter) |
| `search_tables` | Find tables by name or description across libraries |
| `search_columns` | Find columns by name or description across libraries |
| `describe_table` | Get detailed column information |
| `list_views` | List views in a schema (with optional filter) |
| `list_indexes` | List SQL indexes for a table |
| `get_table_constraints` | Get primary keys, foreign keys, unique constraints |
| `list_routines` | List SQL procedures and functions in a library, with language, external program, and SQL data access |
| `describe_routine` | Parameters, return value or result columns, and a call template for a procedure or function |
| `validate_query` | Check a statement without running it, including catalog names |
| `get_object_ddl` | Return the SQL DDL that recreates an object |
| `get_related_objects` | List objects that depend on a table |
| `get_journal_info` | List journal, images, and primary key per table, and flag tables a replication tool cannot read |
| `index_advice` | List the indexes the query optimizer asked for in a library, merged and ranked by temporary index use |
| `profile_table` | Row count, last change, and per-column distinct and null counts from stored statistics or a scan |
| `get_business_context` | List business descriptions, row filters and relations loaded from YAML |
| `search_ibmi_services` | Find IBM i services by keyword or category, with the release that added each one and an example query |
In hosts that support [MCP Apps](https://modelcontextprotocol.io/extensions/apps), results from `execute_query`, `profile_table` and business SQL tools open as a grid and a chart under the tool call. See [Result view](docs/tools.md#result-view).
### Filter Syntax
The list tools support pattern matching:
- `CUST` - Contains "CUST"
- `CUST*` - Starts with "CUST"
- `*LOG` - Ends with "LOG"
## Resources and prompts
Clients that support MCP resources can read a table's context without a tool call, and complete library and table names as you type.
| Resource | Contents | Registered when |
|----------|----------|-----------------|
| `db2i://{schema}/{table}` | Columns from the catalog, plus the YAML business description, column notes, and relations | `describe_table` is enabled |
| `db2i://{schema}/{table}/ddl` | SQL from `QSYS2.GENERATE_SQL` that recreates the table, view, or alias | `get_object_ddl` is enabled |
| `db2i://business-context` | Every annotation loaded from `MCP_CUSTOM_TOOLS` | `get_business_context` is enabled |
| `ui://db2i/result` | The [result view](docs/tools.md#result-view) MCP Apps hosts show under tool results | `MCP_UI_ENABLED` is on (the default) |
`resources/list` offers the annotated tables, for example `db2i://MYLIB/ORDERS`. Percent-encode `#` and other reserved characters in names (`ORD%23X` for `ORD#X`). A library outside `QUERY_ALLOWED_SCHEMAS` is rejected with the same message `execute_query` gives, and completion offers only allowed libraries. Reads and completions that query IBM i count against the rate limit, and reads are written to the audit log. Completion fetches a library's name list once and reuses it for 60 seconds, so typing a name costs one query rather than one per keystroke.
| Prompt | Arguments | What it asks for |
|--------|-----------|------------------|
| `explore_library` | `schema` | List the tables, describe the central ones, and summarize how they join |
| `explain_table` | `schema`, `table` | Explain rows, columns, keys, and relations in plain language |
| `write_query` | `question`, `schema`, `table` | Write one SELECT from the table's real columns and YAML relations, then validate and run it when those tools are enabled |
A prompt is listed only when the tools it tells the model to call are enabled: `explore_library` needs `list_tables` and `describe_table`, and the other two need `describe_table`. None of them asks for a write.
## Use cases
I've used this server on projects where the source system was the Iptor DC1 ERP on IBM i. The same patterns work with any IBM i ERP.
- **Building REST APIs** - The agent finds the ERP tables and keys, checks its SQL with `validate_query`, tests it on sample rows, and then writes the endpoint.
- **ETL and ELT pipelines for BI** - Profile source tables, generate staging DDL with `get_object_ddl`, and draft incremental extracts and code mappings for the warehouse.
- **Near-real-time replication to BI** - Check which tables are journaled, and with which images, before a journal-based tool such as Fivetran streams changes to the warehouse.
- **Ad-hoc analysis** - Connect Claude or Cursor directly to the ERP and ask business questions in plain language, with vetted Business SQL tools and column masking for sensitive fields.
See [Use cases](docs/use-cases.md) for sample prompts and the guardrails that go with each one.
## Example Usage
Once connected, you can ask the AI assistant:
- "List all schemas that contain 'PROD'"
- "Show me the tables in schema MYLIB"
- "Describe the columns in MYLIB/CUSTOMERS"
- "What indexes exist on the ORDERS table?"
- "Run this query: SELECT * FROM MYLIB.CUSTOMERS WHERE STATUS = 'A'"
- "Find the order header and line tables in MYLIB and write a GET /orders/:orderNo endpoint"
- "Draft an incremental extract of MYLIB.ORDERHDR rows changed since yesterday"
## Documentation
| Guide | Description |
|-------|-------------|
| [Tools, resources, and prompts](docs/tools.md) | Built-in tools, filter syntax, MCP resources, and prompts |
| [HTTP Transport](docs/http-transport.md) | HTTP API, auth, and protocol versions |
| [Configuration](docs/configuration.md) | All environment variables and driver options |
| [Security](docs/security.md) | Credentials, rate limiting, query validation |
| [Business SQL tools](docs/custom-tools.md) | YAML tools for orders, ledgers, and master data |
| [Use cases](docs/use-cases.md) | REST APIs, BI pipelines, replication, and ad-hoc analysis |
| [Client Setup](docs/client-setup.md) | Cursor, Claude, Claude Code setup |
| [Example web app](docs/web-app.md) | Browser chat with your own model key and IBM i sign-in |
| [Docker Guide](docs/docker.md) | Container deployment |
| [Development](docs/development.md) | Contributing and local setup |
## Compatibility
- IBM i V7R3 and later (V7R5 recommended)
- `validate_query` and the `execute_query` parse check need `QSYS2.PARSE_STATEMENT` (IBM i 7.3 with Db2 PTF group SF99703 level 3, or 7.4 and later)
- `get_related_objects` needs IBM i 7.3 Technology Refresh 9, IBM i 7.4 Technology Refresh 3, or a later release
- `get_journal_info` needs the journal columns of `QSYS2.OBJECT_STATISTICS` (IBM i 7.3 Technology Refresh 2 or later)
- `search_ibmi_services` needs `QSYS2.SERVICES_INFO`, which ships with the Db2 for i PTF group
- The `cause` and `recovery` on a failed statement come from `SYSTOOLS.SQLCODE_INFO`. Without it, errors return the SQLSTATE, SQLCODE and message only
- Node.js 22 or higher
- unixODBC with the IBM i Access ODBC Driver for the default `odbc` driver, a JDK at install time and a JRE 11 or higher at runtime for the optional `jt400` driver, or SSH access and Java 8 or higher on the IBM i for the optional `mapepire` driver (see [Database Drivers](docs/configuration.md#database-drivers))
- MCP spec 2026-07-28, plus stateless clients from the 2025-era revisions (through 2025-11-25)
## Related Projects
- **[IBM ibmi-mcp-server](https://github.com/IBM/ibmi-mcp-server)** - IBM's official MCP server for IBM i systems. Offers YAML-based SQL tool definitions and AI agent frameworks. Requires [Mapepire](https://mapepire-ibmi.github.io/). This project's `mapepire` driver uses Mapepire's SSH mode, which needs no Mapepire server running on the IBM i.
## Contributing
Contributions are welcome! See the [Development Guide](docs/development.md) for setup instructions.
## License
MIT License - see [LICENSE](LICENSE) for details.
## Trademarks
IBM, IBM i and Db2 are trademarks of International Business Machines Corporation. This project is not affiliated with or endorsed by IBM.
## Acknowledgments
- [node-jt400](https://www.npmjs.com/package/node-jt400) - JT400 JDBC driver wrapper for Node.js
- [node-odbc](https://github.com/IBM/node-odbc) - ODBC bindings for Node.js, maintained by IBM
- [mapepire-js](https://github.com/Mapepire-IBMi/mapepire-js) - Mapepire client for Node.js, maintained by IBM
- [Model Context Protocol](https://modelcontextprotocol.io/) - The protocol specification
- [@modelcontextprotocol/server](https://github.com/modelcontextprotocol/typescript-sdk) - Official TypeScript SDK (spec 2026-07-28, with stateless 2025-era clients)
