@altura-oncology/mcp-server

CureEngine Model Context Protocol server. Exposes a comprehensive cancer research knowledge base as MCP tools for Claude, Cursor, and other AI agents.

Built by Nobre Labs for the CureEngine Cancer Research Foundation.

What's in the knowledge base

• 3.26M+ cancer research papers from PubMed, PMC, Europe PMC, OpenAlex, SciELO, and bioRxiv
• 121K clinical trials from ClinicalTrials.gov and WHO ICTRP
• 8.1M knowledge graph edges from PrimeKG (Harvard Precision Medicine Knowledge Graph)
• 1.96M real patient mutations from cBioPortal (TCGA, MSK-IMPACT, TRACERx)
• 22,577 patient samples with mutation profiles and cancer types
• 1,140 FDA drug records with targets, mechanisms, cancer indications
• 63 medicinal plants with 11,916 natural compounds (SMILES + InChI) from LOTUS
• AI-enriched profiles for cancer drugs, mutations, pathways, supplements, and failed trials

Quick start (hosted mode — recommended)

Get an API key at cureengine.org/api-keys, then add this to your Claude Desktop config:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "cureengine": {
      "command": "npx",
      "args": ["-y", "@altura-oncology/mcp-server"],
      "env": {
        "CUREENGINE_API_KEY": "ce_xxxxxxxxx"
      }
    }
  }
}

Restart Claude Desktop. The cureengine tools will appear in the tool menu.

In Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "cureengine": {
      "command": "npx",
      "args": ["-y", "@altura-oncology/mcp-server"],
      "env": {
        "CUREENGINE_API_KEY": "ce_xxxxxxxxx"
      }
    }
  }
}

Available MCP tools

Hosted mode (CUREENGINE_API_KEY)

| Tool | Description | |------|-------------| | search_papers | Full-text search over 3.26M+ cancer papers with credibility scoring and cancer type filtering | | get_paper | Fetch a single paper by PMID, DOI, or UUID — full abstract, authors, MeSH terms | | search_trials | Search 121K+ clinical trials by keyword, phase, or status | | search_drugs | Find drugs by name, mechanism, target gene, or cancer indication | | match_drugs_to_patient | Given a cBioPortal sample ID, rank drug candidates by mutation match (FDA, off-label, plant compounds) | | explore_gene | Cancer associations, targeting drugs (OncoKB levels), and PrimeKG knowledge graph edges for any gene | | search_plants | Search 63 medicinal plants and 11,916 natural compounds by name, traditional use, or cancer relevance | | find_repurposing_paths | Disease → associated genes → targeting drugs repurposing candidates via PrimeKG |

Direct database mode (DATABASE_URL)

All hosted-mode tools plus:

| Tool | Description | |------|-------------| | get_cancer_overview | Summary stats and top papers/trials for a cancer type | | explore_knowledge_graph | Multi-hop PrimeKG traversal from any node | | get_research_landscape | Research trends, top authors, and publication stats | | run_readonly_sql | Arbitrary SELECT queries over the full schema | | export_csv | Export any query result as a CSV | | find_repurposing_candidates | Expanded drug repurposing with PrimeKG multi-hop paths | | traverse_knowledge_graph | Full PrimeKG traversal with configurable depth and filters | | find_drug_repurposing_paths | Detailed disease → gene → drug path with evidence weights |

Direct database mode

If you have access to a CureEngine Postgres replica, you can run tools directly against the database with no API quota:

{
  "mcpServers": {
    "cureengine": {
      "command": "npx",
      "args": ["-y", "@altura-oncology/mcp-server"],
      "env": {
        "DATABASE_URL": "postgresql://user:password@host:5432/cancer_research"
      }
    }
  }
}

Example queries

Once connected, you can ask Claude (or any MCP-enabled agent):

• "Search for recent papers on KRAS G12C inhibitors in pancreatic cancer"
• "What drugs target BRAF V600E and have Phase 3 evidence?"
• "Match drug candidates for cBioPortal sample TCGA-AA-3562-01"
• "What are the most promising repurposing candidates for glioblastoma?"
• "Show me medicinal plants with compounds targeting NF-kB"
• "Explore the knowledge graph connections for TP53"

Architecture

Two modes, one package:

Hosted mode (CUREENGINE_API_KEY): A thin stdio MCP server that forwards tool calls to the CureEngine HTTP API at cureengine.org. No database setup. API key auth + rate limits apply.

Direct mode (DATABASE_URL): Queries a CureEngine Postgres database directly. Full tool set, no API quota, requires DB access.

The entry point (cli.ts) automatically selects the right mode based on which environment variable is present. If both are set, hosted mode takes precedence.

Each tool:

1. Validates input via Zod schemas
2. Returns markdown-formatted results (narrative) or CSV (tabular)
3. Includes credibility/quality scoring on every result where available

License

MIT

Credits

• Data sources: PubMed, PMC, Europe PMC, ClinicalTrials.gov, WHO ICTRP, cBioPortal, PrimeKG (Harvard), LOTUS (natural products), OncoKB, ChEMBL, OpenFDA
• Built by Nobre Labs for the CureEngine Cancer Research Foundation
• Part of the CureEngine cancer research engine

Support

• Issues: https://github.com/CureEngine/cancer-research-engine/issues
• Email: [email protected]