AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

tamarind

Provides access to a collection of open-source molecular design and structural biology tools on the Tamarind Bio platform, via its REST API or MCP server — no local GPUs required. Tamarind bundles popular open-source models for structure prediction (AlphaFold, Boltz, Chai, ESMFold), protein, binder, and de novo design (RFdiffusion, ProteinMPNN, BoltzGen), antibody and nanobody design and developability, protein-ligand docking (DiffDock, Autodock Vina), binding-affinity prediction, MSA generation, and molecular dynamics. Use when the user mentions Tamarind or tamarind.bio, wants to run any of these open-source tools in the cloud, references app.tamarind.bio/api or the x-api-key header, or needs to submit batches of sequences for structural or biophysical characterization.

Original source
K-Dense-AI
License
MIT
Pinned revision
5f2bc0341354fd5590aef76a533ada1e1e5b4e62

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

Ask your assistant

Use Agentic Memory skills to read k-dense-ai--tamarind at revision 5f2bc0341354fd5590aef76a533ada1e1e5b4e62, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 5

references/examples.md

Tamarind Bio illustrative settings and output interpretation

Reviewed 2026-09-30 using the live public catalog, REST contract, and tool documentation. These examples are not authenticated validate-job results. Replace file placeholders, read GET /tools/{name}/schema, then validate each settings object before a compute submission. Historical exact error messages are not a contract.

Authentication and discovery check

This read-only check does not submit jobs. Supply the user's deployment and key:

import os
import requests

base = os.environ.get("TAMARIND_BASE_URL", "https://app.tamarind.bio/api").rstrip("/")
headers = {"x-api-key": os.environ["TAMARIND_API_KEY"]}
response = requests.get(f"{base}/tools", headers=headers, timeout=(10, 60))
response.raise_for_status()
tools = response.json()
assert isinstance(tools, list)
print("[OK] Account tool catalog returned", len(tools), "entries")

To check a payload next, use /validate-job as shown in workflows. That call requires authentication even though it does not spend a compute job.

AlphaFold monomer and multimer

sequence is the public catalog's required setting for alphafold:

{"sequence": "MQIFVKTLTGKTITLEVEPSDTIENVKAKIQDKEGIPPDQQRLIFAGKQLEDGRTLSDYNIQKESTLHLVLRLRGG"}

For a homodimer, this illustrative short sequence pair shows colon separation:

{"sequence": "MKTAYIAKQRQISFVKSHFSRQLEERLGLIE:MKTAYIAKQRQISFVKSHFSRQLEERLGLIE"}

It does not establish that these chains interact. Read optional model-count, recycle, MSA, and template fields from the account schema instead of assuming string versus numeric representations or defaults.

Boltz and Chai sequence modes

{"inputFormat": "sequence", "sequence": "MKTAYIAKQRQISFVKSHFSRQLEERLGLIEVQAPILSRVGDGTQDNLSGAEKAVQVKVKALP"}

The current public catalog makes inputFormat optional with default "sequence" for both tools. Boltz advertises sequence, list, molecules, and yaml; Chai advertises sequence, molecules, and list (no YAML mode in this catalog). Set a mode explicitly when useful, then supply its conditional fields. Do not claim that omitting inputFormat universally fails validation.

DiffDock with a SMILES ligand

{"proteinFile": "target.pdb", "ligandFormat": "SMILES",
 "ligandSmiles": "CC(=O)Oc1ccccc1C(=O)O"}

Upload an actual target first. The current proteinFile declaration accepts PDB and CIF extensions. ligandFormat defaults to "sdf/mol2 file"; set "SMILES" to activate ligandSmiles. The file branch uses ligandFile with SDF/MOL2.

AutoDock Vina with a SMILES ligand

{"receptorFile": "receptor.pdb", "ligandFormat": "smiles",
 "ligandSmiles": "CC(=O)Oc1ccccc1C(=O)O"}

This is the catalog-backed minimum branch, not a complete scientific docking protocol. Vina uses receptorFile, accepts PDB/CIF, and its ligandFormat values are lowercase "sdf"/"smiles" with default "sdf". Determine a meaningful search box from the receptor/pocket and read its current fields, units, and sampling settings from the full account schema. The public required-settings catalog does not verify optional box parameter names or their defaults; do not reuse coordinates from an unrelated target or claim arbitrary example coordinates are required API settings.

ProteinMPNN inverse folding

{"pdbFile": "backbone.pdb", "designedResidues": {"A": "1 2 3 4 5"}}

The selected residues must exist on chain A of the uploaded structure. The catalog currently declares PDB/CIF input and a space-separated residue-list representation. Its empty-map default does not justify assuming which chains or residues will be redesigned. Inspect the full schema to choose a model variant, sample count, and explicit design region, then validate. Fold generated sequences through the folding tool's sequence field, not a template parameter.

Outputs and failure interpretation

  • Score is optional and can be JSON text, a number, or null; parse by actual type. Metrics and output filenames depend on the tool/task/version.
  • /tools can expose outputs.mainCSV, task-specific outputs.byTask, column metadata, and filterMetrics. Use these and the actual archive to find data; do not assume every folder emits rank_*.pdb or every tool has pLDDT.
  • pLDDT, pTM, and interface confidence are model confidence estimates, not proof of affinity, specificity, or experimental validation. Assess geometries and chemical validity and compare only compatible metrics/scales.
  • Complete permits result retrieval but does not prove usable scientific output. Stopped/Failed require log inspection (fileName: "output.log").
  • /result 202 means archive preparation: wait and repeat retrieval. It is not an instruction to run the scientific job again.
  • /validate-job success returns the normalized settings to submit. An unknown key warning still needs correction; a typo in an optional field can otherwise run with the wrong default.
  • Relative uploaded paths and JobName/path/to/file.ext references are distinct from inline content. Avoid inventing storage keys or treating arbitrary text as an existing file. Current documentation says redundant email prefixes are stripped; older double-prefix errors are no longer universal.