Tamarind Bio
Tamarind runs molecular-design and structural-biology tools on managed compute:
structure prediction, protein and antibody design, docking, binding-affinity
prediction, MSA generation, and molecular dynamics. Use it when the user requests
Tamarind, its REST API/MCP server, or cloud execution of these scientific tools.
For local sequence processing or molecular descriptors, use a local library.
Sources and review scope
The REST contracts and public catalog were reviewed on 2026-09-30. Examples are
illustrative until validated against the user's account; this review did not run
authenticated jobs or establish scientific accuracy for any model.
- API index and
complete guide.
- Current OpenAPI includes discovery,
validation, jobs, files, and newer pipeline/custom-tool surfaces.
openapi.yaml is also available. Check that
needed paths exist: the merged spec can return HTTP 200 with only its classic
surface when the backend spec cannot be fetched.
- Public catalog needs no key; query
?type=
or ?tag=. It documents public tools and conditional required settings, not
every optional parameter or account entitlement.
- Product documentation index links to
Markdown pages, including the MCP guide.
Fetch the account's current schemas before composing a run. Where the prose guide
and OpenAPI differ, prefer the operation/schema for field shapes, and record any
unresolved difference rather than guessing.
Access
- Use the user's Tamarind deployment. The shared base is
https://app.tamarind.bio/api; a dedicated organization deployment has its own
host and account data. Every relative REST path below is under /api.
- Obtain a key from the deployment's API settings and read it from
TAMARIND_API_KEY; send it as x-api-key. Keep keys out of files and logs.
- Check the account's current allowance and compute budget before scaling up.
Free usage is a monthly allowance, not an unconditional promise of ten jobs
forever; billing and entitlements can change.
# Public discovery requires no credential.
curl --fail-with-body 'https://app.tamarind.bio/tools.json?type=alphafold'
# Account-scoped discovery:
curl --fail-with-body 'https://app.tamarind.bio/api/tools' \
-H "x-api-key: $TAMARIND_API_KEY"
For REST examples install requests in the execution environment. The official
CLI distribution is tamarind-cli, and its Custom Tools Python client imports
as from tamarind import Tamarind; the unrelated package named tamarind is not
this client. See the SDK reference.
Core job recipes below use HTTP directly.
Workflow
- Discover. Read
GET /tools and match the user's scientific task to the tool
description. Built-ins return an array; ?custom=true lists legacy custom
tools only. Current custom deployments can be missing from this list: use the
known deployed name and its schema before concluding that it is unavailable.
- Read the schema.
GET /tools/{name}/schema returns a JSON Schema for the
settings object. GET /tools also supplies a trimmed settings parameter
list. Check task-dependent fields, file extensions, list values, and defaults.
- Validate. Send
POST /validate-job with type, settings, and optional
jobName. Check HTTP status first, then JSON valid. On success, inspect and
use normalized as the settings to submit. Address unrecognized_settings
if returned, even alongside valid: true: an optional-field typo can otherwise
silently leave the default in effect. Validation checks fields, not all submit
policies, queue limits, or deployment readiness.
- Submit once.
POST /submit-job takes jobName, type, settings, optional
version for a custom-tool build, and optional/organization-required
projectTag. Persist the submitted name and settings. A successful response
is plain text, not a JSON receipt. Use the returned stored name.
- Poll.
GET /jobs?jobName=... returns a row directly. Single-job terminal
states are Complete, Stopped, and Failed; handle legacy Deleted or an
exact-lookup error without looping indefinitely. Poll batch parents using
batchStatus, and poll newer pipelines on their own run endpoint.
- Download and inspect.
POST /result returns a JSON string URL on 200,
or 202 with status: "preparing". Retry result retrieval after 202, without
resubmitting compute. GET the signed URL without the Tamarind API-key header.
Download the archive only for successful runs; request fileName: "output.log"
for stopped/failed jobs. Verify the scientific outputs after downloading.
Workflow recipes implement validation, stored names,
bounded polling, 202 handling, batch validation, and pagination. They are locally
smoke-tested with simulated responses; authenticated execution remains untested.
Picking tools and interpreting results
Select by inputs, intended output, and modeling assumptions, then confirm the
candidate in the live catalog. These are anchors, not a guaranteed catalog:
| Task |
Candidates and decisions |
| Protein/complex structure |
alphafold for AF2 monomers/multimers; boltz, chai, or protenix for cofolding including ligands/nucleic acids; esmfold for fast single-sequence protein folding. Check esmfold2 separately: its current catalog includes protein, DNA, RNA, and ligand complexes. |
| Binder/motif design |
bindcraft, boltzgen, rfdiffusion; choose by target type, scaffold constraints, and required structure inputs. |
| Inverse folding |
proteinmpnn/ligandmpnn consume structures and design sequences. Re-fold designs and compare to the intended backbone/interface. |
| Small-molecule docking |
autodock-vina for a fixed receptor and search box; diffdock for diffusion docking; boltz/chai for cofolding. Choose the modeling approach for the task, not to avoid supplying a required input. |
| Antibody/developability/MSA/MD |
Filter descriptions and schemas for the specific task; availability and inputs differ by tool. |
Confidence scores describe model confidence, not experimental binding, specificity,
or affinity. Compare designed backbones, interfaces, clashes, chain/residue mapping,
and developability. Check ligand chemistry and stereochemistry; docking scores
are not interchangeable with measured binding free energies. Record tool/model,
input provenance, chain mapping, seeds/samples, MSA/template choices, normalized
settings, and any user-selected filtering thresholds.
Honor the user's selected tool and budget. Use authorized defaults for routine
choices; surface unresolved choices that materially affect the scientific task or
compute scope before a large campaign. Never silently substitute a different
scientific task because its inputs are easier to supply.
File inputs and chaining
- Upload a structure with
PUT /upload/{filename} (binary body; follow the
documented redirect), then reference the registered relative name, e.g.
target.pdb or inputs/target.pdb when ?folder=inputs was used.
- Confirm names using
GET /files; it returns a non-paginated array for the
selected folder, not a list of a job's outputs.
- Prefer these paths over inline file content. The current guide says redundant
account-email prefixes are stripped; there is no longer a universal
double-prefix failure. Arbitrary strings are not necessarily file references.
- Reuse a completed job's file as
JobName/path/to/file.ext, matching the next
parameter's supported extensions and list/scalar shape. Do not guess filenames.
- ProteinMPNN designs must feed a folding tool's sequence field. A structural
template field does not mean "fold this designed sequence". Read generated
FASTA/CSV sequences and validate one folding settings object per sequence.
- Do not author internal fields such as
submit_method, msa, or monomer_msa.
Batches and pipelines
POST /submit-batch accepts one type, a nonempty settings array, batchName,
and optional parallel jobNames. Validate every row using array-mode
/validate-job (up to 1,000 rows per call), and use each row's normalized settings.
Submission allows up to 30,000 expanded jobs, counting design fan-out, and
has a separate approximately 4.5 MB request limit. Split on both constraints.
Do not assume old weightedHoursBudget, maxRuntimeSeconds, or gpuType request
fields enforce a cap: they are absent from the current batch schema. Use confirmed
account controls and an agreed job/sample count.
Poll GET /jobs?jobName=<batchName> until batchStatus is Complete, Stopped,
or AggregationFailed. Subjobs can finish before aggregation. Fetch the archive
through /result and handle 202; resultUrl is optional and is not a reliable
readiness signal. Page GET /jobs?batch=... using startKey to inspect children.
For new saved workflows use the template/run API under /pipelines: read the
current pipeline graph contract, validate the proposed run with
POST /pipelines/validate, submit with POST /pipelines/submit (required
name, bindings, and one of templateId/pipeline), and poll
GET /pipelines/runs/{run_id}. Its statuses are lowercase and separate from job
statuses. The legacy /submit-pipeline and /run-pipeline remain documented;
API reference gives their actual required fields.
MCP alternative
Connect to https://mcp.tamarind.bio/mcp with OAuth 2.1 or the x-api-key header.
The official guide confirms submitJob, submitBatch, getJobs, getResult,
uploadFile, and getFiles. Read the connected server's tools/list schemas
before using signatures or interpreting result envelopes.
If the connection advertises discovery/validation helpers such as
getAvailableTools, getJobSchema, or validateJob, use their current schemas.
Extra helpers, filter vocabularies, submitBatch(fromJob=...), and upload-through-
MCP variants are not guaranteed by the public guide. This review's anonymous
tools/list request returned 401, so their current contracts were not verified.
Use the documented REST equivalents when needed.
Recovery
HTTP auth failures differ by route: classic endpoints can answer 400, jobs
can answer 401 or gateway 403, and usage can answer 401. A 403 is not
proof of a budget error. Check status and the actual response body before changing
settings. Submission errors may be JSON or plain text regardless of Content-Type.
A timeout/5xx on submit does not prove that nothing queued. Look up the persisted
name before retrying; for campaigns use POST /jobs/search with up to 1,000 names
per request. Respect rate limits and avoid one-request-per-job polling at scale.
A 413 rejects the oversized request before creating jobs; split the body or upload
file content separately. DELETE /delete-job is a soft delete: it hides the
job and leaves stored result files intact.
Reference files
- API reference: endpoint shapes, validation,
pagination, authentication differences, and legacy/new pipeline boundaries.
- Tool catalog: schema interpretation and discovery.
- Examples: current catalog-backed settings examples
and tool-specific caveats, explicitly bounded by verification scope.
- Workflows: executable HTTP recipes with local mocked
verification; no authenticated scientific jobs were run for this review.
1---2name: tamarind3description: 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.4license: MIT5compatibility: Requires Python 3.10+, a Tamarind Bio account, and an API key from app.tamarind.bio. Uses the `requests` library against the public REST API (these recipes use HTTP directly). Network access required. Optional MCP server at mcp.tamarind.bio/mcp for agent hosts.6metadata:7 source: https://skillmd.com/skills/lingxling/tamarind8 version: "1.3"9 last-reviewed: "2026-09-30"10 skill-author: Tamarind Bio11 trigger-keywords: protein structure prediction, AlphaFold, Boltz, Chai, ESMFold, protein design, binder design, de novo design, antibody design, nanobody, protein-ligand docking, DiffDock, Autodock Vina, binding affinity, MSA generation, inverse folding, ProteinMPNN, RFdiffusion, BoltzGen, cloud GPU biology, structure prediction API, x-api-key, developability, adme, enzyme, peptide, protein language models, molecular design12 openclaw:13 primaryEnv: TAMARIND_API_KEY14 envVars:15 - name: TAMARIND_API_KEY16 required: true17 description: Tamarind Bio API key sent as the x-api-key header.18---1920# Tamarind Bio2122Tamarind runs molecular-design and structural-biology tools on managed compute:23structure prediction, protein and antibody design, docking, binding-affinity24prediction, MSA generation, and molecular dynamics. Use it when the user requests25Tamarind, its REST API/MCP server, or cloud execution of these scientific tools.26For local sequence processing or molecular descriptors, use a local library.2728## Sources and review scope2930The REST contracts and public catalog were reviewed on **2026-09-30**. Examples are31illustrative until validated against the user's account; this review did not run32authenticated jobs or establish scientific accuracy for any model.3334- [API index](https://app.tamarind.bio/llms.txt) and35 [complete guide](https://app.tamarind.bio/llms-full.txt).36- [Current OpenAPI](https://app.tamarind.bio/api/openapi.json) includes discovery,37 validation, jobs, files, and newer pipeline/custom-tool surfaces.38 [openapi.yaml](https://app.tamarind.bio/openapi.yaml) is also available. Check that39 needed paths exist: the merged spec can return HTTP 200 with only its classic40 surface when the backend spec cannot be fetched.41- [Public catalog](https://app.tamarind.bio/tools.json) needs no key; query `?type=`42 or `?tag=`. It documents public tools and conditional required settings, not43 every optional parameter or account entitlement.44- [Product documentation index](https://docs.tamarind.bio/llms.txt) links to45 Markdown pages, including the [MCP guide](https://docs.tamarind.bio/tamarind/mcp-server.md).4647Fetch the account's current schemas before composing a run. Where the prose guide48and OpenAPI differ, prefer the operation/schema for field shapes, and record any49unresolved difference rather than guessing.5051## Access52531. Use the user's Tamarind deployment. The shared base is54 `https://app.tamarind.bio/api`; a dedicated organization deployment has its own55 host and account data. Every relative REST path below is under `/api`.562. Obtain a key from the deployment's API settings and read it from57 `TAMARIND_API_KEY`; send it as `x-api-key`. Keep keys out of files and logs.583. Check the account's current allowance and compute budget before scaling up.59 Free usage is a monthly allowance, not an unconditional promise of ten jobs60 forever; billing and entitlements can change.6162```bash63# Public discovery requires no credential.64curl --fail-with-body 'https://app.tamarind.bio/tools.json?type=alphafold'65# Account-scoped discovery:66curl --fail-with-body 'https://app.tamarind.bio/api/tools' \67 -H "x-api-key: $TAMARIND_API_KEY"68```6970For REST examples install `requests` in the execution environment. The official71CLI distribution is **`tamarind-cli`**, and its Custom Tools Python client imports72as `from tamarind import Tamarind`; the unrelated package named `tamarind` is not73this client. See the [SDK reference](https://app.tamarind.bio/api-docs/custom-tools-sdk-reference).74Core job recipes below use HTTP directly.7576## Workflow77781. **Discover.** Read `GET /tools` and match the user's scientific task to the tool79 description. Built-ins return an array; `?custom=true` lists legacy custom80 tools only. Current custom deployments can be missing from this list: use the81 known deployed name and its schema before concluding that it is unavailable.822. **Read the schema.** `GET /tools/{name}/schema` returns a JSON Schema for the83 `settings` object. `GET /tools` also supplies a trimmed `settings` parameter84 list. Check task-dependent fields, file extensions, list values, and defaults.853. **Validate.** Send `POST /validate-job` with `type`, `settings`, and optional86 `jobName`. Check HTTP status first, then JSON `valid`. On success, inspect and87 use `normalized` as the settings to submit. Address `unrecognized_settings`88 if returned, even alongside `valid: true`: an optional-field typo can otherwise89 silently leave the default in effect. Validation checks fields, not all submit90 policies, queue limits, or deployment readiness.914. **Submit once.** `POST /submit-job` takes `jobName`, `type`, `settings`, optional92 `version` for a custom-tool build, and optional/organization-required93 `projectTag`. Persist the submitted name and settings. A successful response94 is plain text, not a JSON receipt. Use the returned stored name.955. **Poll.** `GET /jobs?jobName=...` returns a row directly. Single-job terminal96 states are `Complete`, `Stopped`, and `Failed`; handle legacy `Deleted` or an97 exact-lookup error without looping indefinitely. Poll batch parents using98 `batchStatus`, and poll newer pipelines on their own run endpoint.996. **Download and inspect.** `POST /result` returns a JSON string URL on **200**,100 or **202** with `status: "preparing"`. Retry result retrieval after 202, without101 resubmitting compute. GET the signed URL without the Tamarind API-key header.102 Download the archive only for successful runs; request `fileName: "output.log"`103 for stopped/failed jobs. Verify the scientific outputs after downloading.104105[Workflow recipes](references/workflows.md) implement validation, stored names,106bounded polling, 202 handling, batch validation, and pagination. They are locally107smoke-tested with simulated responses; authenticated execution remains untested.108109## Picking tools and interpreting results110111Select by inputs, intended output, and modeling assumptions, then confirm the112candidate in the live catalog. These are anchors, not a guaranteed catalog:113114| Task | Candidates and decisions |115|---|---|116| Protein/complex structure | `alphafold` for AF2 monomers/multimers; `boltz`, `chai`, or `protenix` for cofolding including ligands/nucleic acids; `esmfold` for fast single-sequence protein folding. Check `esmfold2` separately: its current catalog includes protein, DNA, RNA, and ligand complexes. |117| Binder/motif design | `bindcraft`, `boltzgen`, `rfdiffusion`; choose by target type, scaffold constraints, and required structure inputs. |118| Inverse folding | `proteinmpnn`/`ligandmpnn` consume structures and design sequences. Re-fold designs and compare to the intended backbone/interface. |119| Small-molecule docking | `autodock-vina` for a fixed receptor and search box; `diffdock` for diffusion docking; `boltz`/`chai` for cofolding. Choose the modeling approach for the task, not to avoid supplying a required input. |120| Antibody/developability/MSA/MD | Filter descriptions and schemas for the specific task; availability and inputs differ by tool. |121122Confidence scores describe model confidence, not experimental binding, specificity,123or affinity. Compare designed backbones, interfaces, clashes, chain/residue mapping,124and developability. Check ligand chemistry and stereochemistry; docking scores125are not interchangeable with measured binding free energies. Record tool/model,126input provenance, chain mapping, seeds/samples, MSA/template choices, normalized127settings, and any user-selected filtering thresholds.128129Honor the user's selected tool and budget. Use authorized defaults for routine130choices; surface unresolved choices that materially affect the scientific task or131compute scope before a large campaign. Never silently substitute a different132scientific task because its inputs are easier to supply.133134## File inputs and chaining135136- Upload a structure with `PUT /upload/{filename}` (binary body; follow the137 documented redirect), then reference the registered relative name, e.g.138 `target.pdb` or `inputs/target.pdb` when `?folder=inputs` was used.139- Confirm names using `GET /files`; it returns a non-paginated array for the140 selected folder, not a list of a job's outputs.141- Prefer these paths over inline file content. The current guide says redundant142 account-email prefixes are stripped; there is no longer a universal143 double-prefix failure. Arbitrary strings are not necessarily file references.144- Reuse a completed job's file as `JobName/path/to/file.ext`, matching the next145 parameter's supported extensions and list/scalar shape. Do not guess filenames.146- ProteinMPNN designs must feed a folding tool's **sequence** field. A structural147 template field does not mean "fold this designed sequence". Read generated148 FASTA/CSV sequences and validate one folding settings object per sequence.149- Do not author internal fields such as `submit_method`, `msa`, or `monomer_msa`.150151## Batches and pipelines152153`POST /submit-batch` accepts one `type`, a nonempty `settings` array, `batchName`,154and optional parallel `jobNames`. Validate **every** row using array-mode155`/validate-job` (up to 1,000 rows per call), and use each row's normalized settings.156Submission allows up to **30,000 expanded jobs**, counting design fan-out, and157has a separate approximately **4.5 MB** request limit. Split on both constraints.158Do not assume old `weightedHoursBudget`, `maxRuntimeSeconds`, or `gpuType` request159fields enforce a cap: they are absent from the current batch schema. Use confirmed160account controls and an agreed job/sample count.161162Poll `GET /jobs?jobName=<batchName>` until `batchStatus` is `Complete`, `Stopped`,163or `AggregationFailed`. Subjobs can finish before aggregation. Fetch the archive164through `/result` and handle 202; `resultUrl` is optional and is not a reliable165readiness signal. Page `GET /jobs?batch=...` using `startKey` to inspect children.166167For new saved workflows use the template/run API under `/pipelines`: read the168current pipeline graph contract, validate the proposed run with169`POST /pipelines/validate`, submit with `POST /pipelines/submit` (required170`name`, `bindings`, and one of `templateId`/`pipeline`), and poll171`GET /pipelines/runs/{run_id}`. Its statuses are lowercase and separate from job172statuses. The legacy `/submit-pipeline` and `/run-pipeline` remain documented;173[API reference](references/api_reference.md) gives their actual required fields.174175## MCP alternative176177Connect to `https://mcp.tamarind.bio/mcp` with OAuth 2.1 or the `x-api-key` header.178The official guide confirms `submitJob`, `submitBatch`, `getJobs`, `getResult`,179`uploadFile`, and `getFiles`. Read the connected server's `tools/list` schemas180before using signatures or interpreting result envelopes.181182If the connection advertises discovery/validation helpers such as183`getAvailableTools`, `getJobSchema`, or `validateJob`, use their current schemas.184Extra helpers, filter vocabularies, `submitBatch(fromJob=...)`, and upload-through-185MCP variants are not guaranteed by the public guide. This review's anonymous186`tools/list` request returned 401, so their current contracts were not verified.187Use the documented REST equivalents when needed.188189## Recovery190191HTTP auth failures differ by route: classic endpoints can answer **400**, jobs192can answer **401** or gateway **403**, and usage can answer **401**. A 403 is not193proof of a budget error. Check status and the actual response body before changing194settings. Submission errors may be JSON or plain text regardless of Content-Type.195196A timeout/5xx on submit does not prove that nothing queued. Look up the persisted197name before retrying; for campaigns use `POST /jobs/search` with up to 1,000 names198per request. Respect rate limits and avoid one-request-per-job polling at scale.199A 413 rejects the oversized request before creating jobs; split the body or upload200file content separately. `DELETE /delete-job` is a **soft delete**: it hides the201job and leaves stored result files intact.202203## Reference files204205- [API reference](references/api_reference.md): endpoint shapes, validation,206 pagination, authentication differences, and legacy/new pipeline boundaries.207- [Tool catalog](references/tool_catalog.md): schema interpretation and discovery.208- [Examples](references/examples.md): current catalog-backed settings examples209 and tool-specific caveats, explicitly bounded by verification scope.210- [Workflows](references/workflows.md): executable HTTP recipes with local mocked211 verification; no authenticated scientific jobs were run for this review.