pyomop

A Python package for working with OHDSI OMOP Common Data Model databases that ships an MCP server, so Claude can list CDM tables, inspect their columns, and run validated SQL against a local or remote CDM instance.


Type	MCP server (bundled in a Python package)
Supplier	Bell Eapen (open source)
Availability	GA — PyPI `pyomop` 6.4.0; repo last pushed 2026-07-28
Pricing	Free / OSS — GPL-3.0 per the GitHub license field. Copyleft: redistributing a derived pipeline carries obligations.
Capabilities	Read/Write — SQL execution plus CDM schema creation, against a database you point it at
Verified	degraded · 2026-08-06 — pyomop-mcp-server boot-errored in smoke test; fixed to the working pyomop –mcp-server form
Security	cleared · 2026-08-06 — GPL-3.0 confirmed via direct LICENSE fetch, copyleft only

How to install

Requires Python 3.11+ (requires_python: >=3.11,<4.0). Install the package first, then register the server.

Install:
```
pip install pyomop
```
Add pip install "pyomop[llm]" for the LangChain-backed plain-text query path, or pip install "pyomop[http]" if you want the HTTP transport.
Verify it starts (one-shot check — Ctrl-C once it boots; Claude Code and Claude Desktop launch the process themselves over stdio):
```
pyomop --mcp-server
```

Claude Code — stdio:

claude mcp add --transport stdio pyomop -- pyomop --mcp-server

Claude Desktop — add to claude_desktop_config.json:
```
{
  "mcpServers": {
    "pyomop": {
      "command": "pyomop",
      "args": ["--mcp-server"]
    }
  }
}
```
(A clean-install smoke test this run found the previously-documented pyomop-mcp-server console script absent from PATH after pip install pyomop, even though the package and its mcp dependency installed cleanly — so this page now points at the pyomop --mcp-server subcommand, which is the form the upstream README itself uses via uv run pyomop --mcp-server.)
HTTP transport (optional; a long-lived service you must keep running in another terminal):
```
pyomop-mcp-server-http --host 0.0.0.0 --port 8000
```

What it does

Exposes eight tools over a SQLAlchemy connection to an OMOP CDM v5.4/v6 database:

Schema inspection — get_usable_table_names, get_table_columns, get_single_table_info.
Querying — run_sql, check_sql (validate a statement before executing it), example_query.
Setup — create_cdm (build the CDM schema) and create_eunomia (load the OHDSI Eunomia demo dataset, useful for trying the server without patient data).

Backends: SQLite, PostgreSQL, MySQL. The package also converts result sets to pandas DataFrames and includes FHIR import utilities and an agent-assisted ETL migration CLI (pyomop-migrate) outside the MCP surface.

Primary use cases: exploratory querying of an OMOP CDM instance, cohort-count sanity checks, learning the CDM schema against the Eunomia demo database.

Notes

create_cdm and create_eunomia are deliberately restricted to local SQLite databases upstream, to avoid destructive operations against a production CDM. run_sql is not so restricted — point this server at a real clinical database only with a read-only database role and the usual governance review, since the model composes the SQL.

The MCP server covers CDM data access; it does not bundle the OMOP vocabularies. Upstream suggests pairing it with a vocabulary service when Athena vocabularies are not loaded locally — see OMOPHub MCP Server.

GPL-3.0 confirmed by fetching the repository’s LICENSE file directly from the develop branch (the repo’s default) this run — the PyPI metadata still leaves license_expression empty, but the rendered file matches the GitHub license field verbatim.

Sources

Installed this tool?

Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.