> ## Documentation Index
> Fetch the complete documentation index at: https://docs.biohub.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Biohub MCP example prompts for ESM Atlas research

> Copy example prompts that use the Biohub MCP server to find similar proteins, resolve accessions, explain SAE features, and view Mol* structures.

These prompts work in any client connected to Biohub MCP.
Each one lists the tools the assistant typically calls and what came back when we ran it.
Your assistant's wording will differ, and ESM Atlas scores can shift as the data is updated.

<Tip>
  Include the sequence or an accession in your prompt when you can.
  A protein name can match many UniProt records, and the ESM Atlas only matches the exact sequence you send.
</Tip>

Two proteins appear throughout.
GB1 is the 56-residue B1 domain of protein G:

```text theme={null}
MTYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDDATKTFTVTE
```

GFP is green fluorescent protein from *Aequorea victoria*, UniProt [P42212](https://www.uniprot.org/uniprotkb/P42212).

## Find similar proteins

### Find the nearest ESM Atlas neighbors of GB1

<Prompt description="Find the five ESM Atlas proteins most similar to GB1 (sequence above). For each, give the accession, similarity score, cluster size, and protein name.">
  Find the five ESM Atlas proteins most similar to GB1: MTYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDDATKTFTVTE.
  For each, give the accession, similarity score, cluster size, and protein name.
</Prompt>

**Tools:** `esm_atlas_search_similar_protein_clusters`.

**What to expect:** five hypothetical proteins, mostly from marine metagenomes, 61 to 64 residues long, with similarity scores of about 0.81 to 0.82 and clusters of 50 to 72 members.
All five share GB1's two strongest features, 2822 and 4328.

### Find uncharacterized clusters similar to GFP

<Prompt description="Look up UniProt P42212, then find three ESM Atlas clusters similar to it that have no characterized Pfam domains. List their accessions, lengths, and similarity scores.">
  Look up UniProt P42212, then find three ESM Atlas clusters similar to it that have no characterized Pfam domains.
  List their accessions, lengths, and similarity scores.
</Prompt>

**Tools:** `esm_atlas_search_uniprot`, then `esm_atlas_search_similar_protein_clusters` with `uncharacterized_only` set to `true`.

**What to expect:** three hypothetical proteins, 97 to 334 residues long, with similarity scores of about 0.58 to 0.60.
These are weaker matches than the GFP-domain cluster, which the filter excludes because it is fully characterized.

## Resolve an accession

### Resolve a UniParc accession

<Prompt description="Resolve UniParc accession UPI0000002FB4 and tell me whether its sequence matches UniProt P42212.">
  Resolve UniParc accession UPI0000002FB4 and tell me whether its sequence matches UniProt P42212.
</Prompt>

**Tools:** `esm_atlas_lookup_accession`, then `esm_atlas_search_uniprot`.

**What to expect:** a 238-residue UniParc sequence identical to GFP.

### Find human CDK2 with a fielded UniProt query

<Prompt description="Search UniProt for gene:CDK2 AND organism_id:9606 AND reviewed:true and report the accession and length.">
  Search UniProt for gene:CDK2 AND organism\_id:9606 AND reviewed:true and report the accession and length.
</Prompt>

**Tools:** `esm_atlas_search_uniprot`.

**What to expect:** one result, P24941, Cyclin-dependent kinase 2 from *Homo sapiens*, 298 residues.

## Explain SAE features

### Profile GB1's strongest SAE features

<Prompt description="What are the strongest ESMC SAE features for GB1 (sequence above)? For each, give the label, how reliable the label is, and which residues drive it.">
  What are the strongest ESMC SAE features for GB1: MTYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDDATKTFTVTE?
  For each, give the label, how reliable the label is, and which residues drive it.
</Prompt>

**Tools:** `esm_atlas_get_protein_details`.

**What to expect:** ten features.
The strongest is feature 2822, "Central helix-strand acidic-hydrophobic motif", with moderate reliability and a region at positions 34 to 55.
Next come 4328, "Charged amphipathic recognition helix", and 1089, "N-terminal leader docking helix".
GB1 is already in the ESM Atlas, so these come from the catalog rather than being computed.
Region positions start at 0, so position 34 is residue 35.

### Explain one SAE feature

<Prompt description="Explain ESM Atlas SAE feature 2822. Which SwissProt proteins activate it most, and which features are its nearest neighbors?">
  Explain ESM Atlas SAE feature 2822.
  Which SwissProt proteins activate it most, and which features are its nearest neighbors?
</Prompt>

**Tools:** `esm_atlas_get_sae_feature_detail`.

**What to expect:** a "Compositional bias" feature that marks a short central motif mixing hydrophobic and acidic residues, often where a helix turns into a strand in small compact proteins.
The top SwissProt activation is P33230, and the nearest decoder features are 1255, 5928, 15069, 4746, and 1293.

## Explore cluster context

### Summarize the ESM Atlas cluster closest to GFP

<Prompt description="Find the ESM Atlas cluster closest to GFP (UniProt P42212) and summarize it: how big it is, how well characterized it is, its Pfam domains, and its most common phyla.">
  Find the ESM Atlas cluster closest to GFP (UniProt P42212) and summarize it: how big it is, how well characterized it is, its Pfam domains, and its most common phyla.
</Prompt>

**Tools:** `esm_atlas_search_uniprot`, then `esm_atlas_search_similar_protein_clusters`, then `esm_atlas_get_cluster_info` on the top hit.

**What to expect:** the top hit is a 239-residue "Green fluorescent protein domain-containing protein" with a similarity score of about 0.90.
Its cluster has 64 members, all characterized, and all 64 carry Pfam domain PF01353 (Green fluorescent protein).
Cnidaria is the most common phylum.

## View a structure

### Show GB1's structure

<Prompt description="Show me the structure of GB1 (sequence above).">
  Show me the structure of GB1: MTYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDDATKTFTVTE.
</Prompt>

**Tools:** `ui_show_protein_structure`.

**What to expect:** the ESM Atlas has no stored structure for GB1, so the server predicts one, which took about 20 seconds.
The result is marked as a prediction folded on a miss, with 56 residues and 435 atoms: a four-stranded beta sheet packed against one helix.
Clients that support MCP Apps show the interactive viewer, and others show the PNG preview or the text result.

### Color GB1 by confidence and highlight W43

<Prompt description="Show GB1 colored by confidence, with Trp43 on chain A highlighted and turned toward me.">
  Show GB1 colored by confidence, with Trp43 on chain A highlighted and turned toward me.
</Prompt>

**Tools:** `ui_show_protein_structure` with these view options:

```json theme={null}
{
  "color_by": "confidence",
  "orientation": "hotspot-facing",
  "highlights": [
    { "chain_id": "A", "author_residue_number": 43, "expected_residue": "W" }
  ]
}
```

**What to expect:** a cartoon colored mostly in the high-confidence blues, with Trp43 drawn in the highlight color and facing the camera.
This call predicts the structure again, so its coordinates can differ slightly from the previous view.

### Show a stored ESM Atlas structure

<Prompt description="Find the ESM Atlas cluster closest to GFP (UniProt P42212), show me the structure of its representative, and give its pTM and mean pLDDT.">
  Find the ESM Atlas cluster closest to GFP (UniProt P42212), show me the structure of its representative, and give its pTM and mean pLDDT.
</Prompt>

**Tools:** `esm_atlas_search_uniprot`, `esm_atlas_search_similar_protein_clusters`, `ui_show_protein_structure` with the top hit's sequence, and `esm_atlas_get_protein_details` for the confidence scores.

**What to expect:** a structure that was already in the ESM Atlas (`source.kind` is `catalog`, so nothing was predicted for this request), with 239 residues and 1,897 atoms.
Its ESM Atlas record reports a pTM of about 0.94 and a mean pLDDT of about 0.95.

## Next steps

* Look up each tool's inputs and outputs in the [ESM Atlas tools](/biohub-mcp/esm-atlas-tools) reference.
* Learn what the viewer can do in [Structure viewer](/biohub-mcp/structure-viewer).
* Read [Security and privacy](/biohub-mcp/security) before sending sequences you consider sensitive.