Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@mcp-bindingdb What's the binding affinity of imatinib for ABL1?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Under construction: this project is in early development. Tools, table layouts and examples may change without notice.
An MCP server for querying BindingDB (about 3.2M measured protein–ligand binding affinities) from a local, read-only DuckDB build of the monthly MySQL dump.
Requires Docker (daemon running) and uv.
scripts/build_duckdb.sh # release 202610 -> data/bindingdb_202610.duckdb
scripts/build_duckdb.sh --release 202611 # a later releaseThe script downloads the dump, loads it into a temporary MySQL 8.4 container, copies it to DuckDB
(scripts/convert_to_duckdb.py, checking row counts), adds derived query tables
(scripts/add_derived_tables.py), and removes the container. The regusers and person tables
(passwords and contact details) are left out unless you pass --include-pii.
Derived tables:
Table | Contents |
| One row per measured value: |
| One row per small molecule: preferred name, InChIKey, SMILES, formula, weight, counts |
| Compound synonyms, including ChEMBL ids, PubChem |
| One row per polymer/complex target: bare UniProt accession, organism, counts |
| Target synonyms (gene names, UniProt entry names) |
Related MCP server: guidetopharmacology
uv run mcp-bindingdb # stdio
uv run mcp-bindingdb --transport streamable-http --port 8000The database is the newest data/bindingdb_*.duckdb, or --db PATH / $BINDINGDB_DUCKDB.
Claude Code: .mcp.json in this repo registers the server for the project.
Claude Desktop: add to claude_desktop_config.json:
{
"mcpServers": {
"mcp-bindingdb": {
"command": "uv",
"args": ["--directory", "/path/to/mcp-bindingdb", "run", "mcp-bindingdb"]
}
}
}Tool | Purpose |
| By name/synonym, ChEMBL id, |
| By name, gene name, UniProt entry name or accession; optional organism filter |
| Identifiers, synonyms, PDB ligands, most potent targets |
| Identifiers, synonyms, PDB ids, complex components, measurement counts |
| Most potent compounds for a target (by id or UniProt accession, optional variants) |
| Activity/selectivity profile of a compound |
| Full record of one measurement, with assay description and citation |
| Table list with descriptions, or columns and sample rows of one table |
| Arbitrary read-only DuckDB SQL |
All tools are read-only. The connection is opened read-only with file, network and extension access
disabled and settings locked, so run_sql cannot write data or read anything outside the database.
Queries time out after 30 s (--timeout).
Ki/Kd/IC50/EC50 are in nM, kon in M⁻¹s⁻¹, koff in s⁻¹; p_affinity = 9 − log10(nM).
> values (e.g. >10000) usually mean inactive at the highest concentration tested; potency filters exclude them.
monomer.display_name holds only the BDBM id; names are in compound_name / compound.name.
monomer.chembl_id is empty in this release; ChEMBL ids appear as synonyms instead.
Mutant and construct targets are separate rows sharing a UniProt accession (uniprot_raw like P00533[L858R]).
Text comparisons in DuckDB are case-sensitive (MySQL's were not): use ILIKE or lower().
examples/ has twelve transcripts of real questions answered with these tools: each one shows the tool
calls, the results, and an answer. Regenerate them after a new release with
uv run python scripts/generate_examples.py. Hand-written answers are kept, and the script lists any
example whose results have changed so its answer can be reviewed.
uv run pytestBSD 3-Clause; see LICENSE. BindingDB data are subject to BindingDB's own terms.
This server cannot be deployed
BindingDB — measured protein/small-molecule binding affinities.
Link compounds to protein targets, rank bioactivity, and look up drug mechanisms and indications.
Query, join, profile, clean and convert CSV/JSON/Parquet with server-side DuckDB over MCP.
Biomedical data: compounds, drug info, and molecular targets