opentargets-py

Modern Python client for the Open Targets Platform GraphQL API — with sync and async clients, a --json CLI, and an MCP server so AI agents can query it directly.

Disclaimer: Unofficial, community-maintained. Not affiliated with or endorsed by the Open Targets consortium. For the official platform, visit platform.opentargets.org.

The official opentargets package was deprecated when Open Targets migrated to GraphQL in 2021 and has since been removed from PyPI. This library fills that gap — it is the only Python SDK targeting the current GraphQL API.

Installation

pip install opentargets-py                 # core SDK (sync + async)
pip install opentargets-py[pandas]         # adds DataFrame output
pip install opentargets-py[cli]            # adds the `opentargets` CLI
pip install opentargets-py[mcp]            # adds the `opentargets-mcp` MCP server
pip install opentargets-py[all]            # everything

Quick start

Sync

from opentargets import OpenTargetsClient

with OpenTargetsClient() as client:
    target = client.get_target("EGFR")
    print(target.approved_name)  # epidermal growth factor receptor

    associations = client.get_target_associations("EGFR", limit=10)
    for a in associations:
        print(a.disease_name, round(a.score, 3))

Async (concurrent fan-out)

import asyncio
from opentargets import AsyncOpenTargetsClient

async def main():
    async with AsyncOpenTargetsClient() as client:
        targets = await asyncio.gather(
            client.get_target("EGFR"),
            client.get_target("BRAF"),
            client.get_target("KRAS"),
        )
        for t in targets:
            print(t.approved_symbol, t.biotype)

asyncio.run(main())

CLI

opentargets target EGFR                          # rich table
opentargets target EGFR --json                   # machine-parseable
opentargets targets EGFR BRAF KRAS --json        # batch
opentargets disease EFO_0000311 --targets --json | jq '.targets[0]'
opentargets search "lung cancer" --json

Every subcommand supports --json, so the CLI is usable from shell scripts, agents, and notebooks alike.

MCP server (Claude Desktop, Cursor, etc.)

pip install opentargets-py[mcp]   # requires Python 3.10+

Then add to your claude_desktop_config.json (macOS path: ~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "opentargets": {
      "command": "opentargets-mcp"
    }
  }
}

Now you can ask Claude things like "What diseases is EGFR associated with?" and it will call the SDK directly. Exposed tools: get_target_info, find_target_associations, get_target_drugs, get_target_tractability, get_target_safety, get_target_expression, get_target_constraint, get_disease_info, find_disease_targets, get_drug_info, get_drug_indications, search_open_targets.

Features

• Sync + async clients — OpenTargetsClient and AsyncOpenTargetsClient with full feature parity
• 15 endpoint methods — targets, diseases, drugs, associations, tractability, safety liabilities, baseline expression, genetic constraint, ChEMBL crosslinks, search
• Symbol resolution — pass "EGFR" instead of "ENSG00000146648"
• Auto-pagination — fetches all pages transparently
• Caching — in-memory TTLCache by default; opt into DiskCache (SQLite, survives process restarts) for repeated runs
• Configurable retries — RetryConfig(max_retries=..., base_delay=..., retryable_statuses=..., respect_retry_after=...)
• Pandas integration — as_dataframe=True on list-returning methods
• Typed — full Pydantic v2 models, py.typed marker, mypy --strict compliant
• CLI — opentargets with --json on every command, rich tables for humans
• MCP server — opentargets-mcp, drop-in for Claude Desktop / Cursor / any MCP host
• Minimal core — just httpx + pydantic; everything else is an opt-in extra

Configuration

Disk cache (persist across runs)

from opentargets import OpenTargetsClient, DiskCache

cache = DiskCache(path="~/.cache/opentargets.db", ttl=86400)  # 1 day
with OpenTargetsClient(cache=cache) as client:
    target = client.get_target("EGFR")  # first run: API
    target = client.get_target("EGFR")  # second run (even new process): cache hit

Custom retry policy

from opentargets import OpenTargetsClient, RetryConfig

retry = RetryConfig(max_retries=5, base_delay=2.0, max_delay=120.0)
client = OpenTargetsClient(retry_config=retry)

For AI agents

This package was designed to be usable by both humans and LLM-based agents. Three deliberate affordances:

1. MCP server (opentargets-mcp) — wraps every public method as an MCP tool with LLM-friendly docstrings. Configure once and the model calls Open Targets directly. See MCP setup above.
2. CLI with --json — every subcommand emits clean JSON, so agents can shell out and parse the result with jq or json.loads.
3. Typed Pydantic responses — every method returns a model with .model_dump() / .model_dump_json(), so programmatic agents get structured data without parsing.

See llms.txt for a skimmable one-page overview of the full API surface, optimized for LLM ingestion.

More examples

See the examples/ directory:

• basic_usage.py — targets, diseases, drugs, search
• batch_query.py — batch fetch + DataFrame output
• network_analysis.py — build a target–disease network

API reference

See docs/api-reference.md. All public methods on OpenTargetsClient (and the matching AsyncOpenTargetsClient) have docstrings with Args, Returns, and Example blocks.

Contributing

1. Fork the repo and create a feature branch.
2. Install dev dependencies: pip install -e ".[dev,all]"
3. Run tests: pytest
4. Run linting + formatting: ruff check src tests && ruff format src tests
5. Type check: mypy src/opentargets
6. Open a pull request.

License

Apache 2.0 — see LICENSE.