# Biohub Platform

> Developer docs for Biohub: the ESMC, ESMFold2, and ESM3 protein models, the ESM Atlas API, and the Biohub MCP server at https://biohub.ai/mcp.

- [ESMC](https://docs.biohub.ai/models/esmc.md): ESMC is the latest in the ESM family of protein language models, establishing a new frontier in representation learning for protein biology.
- [ESMFold2](https://docs.biohub.ai/models/esmfold2.md): ESMFold2 is the successor to ESMFold that sets a new state of the art for single-sequence structure prediction.
- [ESM3](https://docs.biohub.ai/models/esm3.md): ESM3 is a multimodal generative protein language model that jointly models sequence, structure, and function.
- [Get Started with ESM Models](https://docs.biohub.ai/learn/getting-started.md): Set up the esm Python package and choose from ESMC, ESMFold2, SAEs, or ESM3 to begin fine-tuning or working with state-of-the-art biological AI models.
- [FAQ](https://docs.biohub.ai/learn/faq.md): Answers to common questions about the Biohub Platform, including open-source models, datasets, and research resources.
- [ESMC](https://docs.biohub.ai/learn/tutorials/esmc.md): Embed sequences, score mutations, interpret SAE features, and fine-tune ESMC.
- [Embed multiple sequences with ESMC](https://docs.biohub.ai/learn/tutorials/esmc/batch-embed-esmc.md): Use the esm SDK parallel executor to generate ESMC representations for multiple protein sequences.
- [ESMFold2](https://docs.biohub.ai/learn/tutorials/esmfold2.md): Fold proteins with DNA, RNA, and ligands, and design binders with ESMFold2.
- [ESM3](https://docs.biohub.ai/learn/tutorials/esm3.md): Represent, generate, and guide the design of novel proteins with ESM3.
- [Atlas API guides](https://docs.biohub.ai/learn/tutorials/atlas/overview.md): Complete common ESM Atlas tasks with public REST endpoints.
- [Search for similar proteins](https://docs.biohub.ai/learn/tutorials/atlas/search-similar-proteins.md): Find Atlas proteins with SAE feature profiles similar to a query amino-acid sequence.
- [Resolve a UniProt accession](https://docs.biohub.ai/learn/tutorials/atlas/resolve-uniprot.md): Resolve a UniProtKB accession or entry name to sequence, metadata, and an Atlas protein hash.
- [Retrieve protein and cluster context](https://docs.biohub.ai/learn/tutorials/atlas/retrieve-protein-cluster.md): Resolve a UniProt accession, retrieve its Atlas protein record, and inspect its feature-based cluster.
- [Browse SAE features](https://docs.biohub.ai/learn/tutorials/atlas/browse-sae-features.md): List the ESMC SAE feature catalog and inspect one feature's supporting metadata.
- [ESM Atlas](https://docs.biohub.ai/learn/guides/esm-atlas.md): Explore the global protein universe with predicted 3D structures from ESMFold2 and interpretable SAE feature annotations from ESMC.
- [Biohub MCP Server](https://docs.biohub.ai/biohub-mcp/overview.md): Connect ChatGPT, Codex, Claude, or Cursor to the Biohub MCP server to search the ESM Atlas, explain protein SAE features, and view Mol* structures.
- [Connect to Biohub MCP and run your first query](https://docs.biohub.ai/biohub-mcp/quickstart.md): Add the Biohub MCP server to Codex, Claude, ChatGPT, or Cursor in a few minutes, confirm its seven tools, and run a first protein lookup.
- [Connect Codex, Claude, ChatGPT, or Cursor to Biohub MCP](https://docs.biohub.ai/biohub-mcp/connect-clients.md): Set up the Biohub MCP server in Codex, Claude Code, Claude.ai, Claude Desktop, ChatGPT developer mode, Cursor, VS Code, or any other MCP client.
- [Biohub MCP example prompts for ESM Atlas research](https://docs.biohub.ai/biohub-mcp/example-prompts.md): Copy example prompts that use the Biohub MCP server to find similar proteins, resolve accessions, explain SAE features, and view Mol* structures.
- [Biohub MCP ESM Atlas tools reference](https://docs.biohub.ai/biohub-mcp/esm-atlas-tools.md): Reference for the six Biohub MCP ESM Atlas tools: UniProt search, accession lookup, protein details, similarity search, cluster info, and SAE features.
- [Biohub MCP structure viewer: Mol* protein views](https://docs.biohub.ai/biohub-mcp/structure-viewer.md): Show a protein structure with the Biohub MCP ui_show_protein_structure tool, set view_options, and learn what the Mol* app and PNG preview return.
- [Biohub MCP security, privacy, and data handling](https://docs.biohub.ai/biohub-mcp/security.md): Use the Biohub MCP server safely: treat tool output as untrusted, know which services receive your protein queries, and read results as predictions.
- [Troubleshoot Biohub MCP connections and tool errors](https://docs.biohub.ai/biohub-mcp/troubleshooting.md): Fix common Biohub MCP problems such as HTTP 405, the wrong transport, missing tools, and timeouts, and look up every tool error code and limit.
- [Response formats and payload size](https://docs.biohub.ai/api/protein/models/response-formats.md): Choose JSON or binary responses for protein model APIs.
- [logits](https://docs.biohub.ai/api/protein/models/logits.md): (ESM3, ESMC) Performs one inference step and returns logits, embeddings, and SAE features (ESMC only).
- [encode](https://docs.biohub.ai/api/protein/models/encode.md): (ESM3, ESMC) Tokenize sequence, structure, or function annotations
- [decode](https://docs.biohub.ai/api/protein/models/decode.md): (ESM3, ESMC) Decodes tokens sequence, structure, or function annotations
- [generate](https://docs.biohub.ai/api/protein/models/generate.md): (ESM3) Generates an output track conditioned on inputs
- [generate_tensor](https://docs.biohub.ai/api/protein/models/generate_tensor.md): (ESM3) Generates an output track conditioned on inputs. Accepts and returns raw tokens.
- [forward_and_sample](https://docs.biohub.ai/api/protein/models/forward_and_sample.md): (ESM3) Performs one inference step and samples output tokens
- [fold](https://docs.biohub.ai/api/protein/models/fold.md): (ESM3, ESMFold2) Folds proteins. Defaults to esmfold2-fast-2026-05 if no model is given
- [fold_all_atom](https://docs.biohub.ai/api/protein/models/fold_all_atom.md): (ESMFold2) Folds molecular complexes containing proteins, dna, rna, and ligands. Defaults to esmfold2-fast-2026-05 if no model is given
- [inverse_fold](https://docs.biohub.ai/api/protein/models/inverse_fold.md): (ESM3) Inverse folds proteins. Defaults to esm3-open-2024-03 if no model is given
- [Get cluster info and member hashes for a representative protein](https://docs.biohub.ai/api/protein/atlas/clusters.md): Look up cluster metadata and member protein hashes for a cluster representative protein identified by its MD5 hash.
- [Search for similar proteins using SAE feature vectors](https://docs.biohub.ai/api/protein/atlas/similarity-search.md): Submit an amino acid sequence and find similar proteins in the atlas based on SAE feature vector similarity.
- [Get detailed metadata for a single SAE feature](https://docs.biohub.ai/api/protein/atlas/features/detail.md): Return all available metadata for one SAE feature: longform description, top activating UniRef90 and SwissProt proteins, decoder nearest neighbors, and activation statistics.
- [Get Features](https://docs.biohub.ai/api/protein/atlas/features/list.md): Return all SAE features (feature_index, label, description).
- [Get protein metadata, sequence, and top SAE features](https://docs.biohub.ai/api/protein/atlas/proteins/get.md): Look up a protein by its MD5 hash and return its metadata, amino acid sequence, and top-activating SAE features.
- [Get a protein structure thumbnail PNG](https://docs.biohub.ai/api/protein/atlas/proteins/thumbnail.md): Stream a pre-rendered protein structure thumbnail PNG. Sub-resource of /proteins/{hash}. The PNG is content-addressed by hash, so the response is cached aggressively (immutable, 1 year).
- [Resolve a UniProtKB accession to a protein record](https://docs.biohub.ai/api/protein/atlas/proteins/uniprot.md): Look up a UniProtKB accession or entry name (e.g. 'P24941', 'P53_HUMAN') and return its protein record: name, gene, organism, function, sequence, and the Atlas `protein_hash`. UniProtKB only — other accession namespaces (UniParc, MGnify, IMG) return 400.
- [Batch protein lookup](https://docs.biohub.ai/api/protein/atlas/proteins/batch.md): Look up multiple proteins by hash. Small batches return a zip file with per-data-type files (200). Large batches return a job handle for async polling (202).
- [Cancel an async batch protein lookup job](https://docs.biohub.ai/api/protein/atlas/proteins/batch-cancel.md): Request cancellation of an async batch protein lookup job. Cancellation is idempotent: returns 204 whether the job is still pending, already complete, or already cancelled. A job that has already produced results keeps them; subsequent GETs on the job will return the original outcome. Returns 404 if…
- [Poll batch protein lookup job status](https://docs.biohub.ai/api/protein/atlas/proteins/batch-status.md): Check the status of an async batch protein lookup. Returns 200 with a zip download URL when complete, 202 when pending, 410 when expired.

## OpenAPI Specs

- [biohub-platform-openapi](/openapi/biohub-platform-openapi.json)
- [metagenomic-atlas-openapi](/openapi/metagenomic-atlas-openapi.json)