STRING Database

Use the STRING MCP tools as the source of truth for protein interaction, network, enrichment, annotation, evidence, species, and homolog questions.

Grounding Rules

• Use STRING tool output for factual claims about interactions, functions, pathways, enrichment, annotations, homologs, and species availability.
• If no species is provided, use human, NCBI taxon ID 9606.
• Do not guess, assemble, or derive STRING links or image URLs. Only show exact links and image URLs returned by tools.
• If a requested claim is not supported by the current tool output in context, retrieve the relevant STRING data with parameters specific to the new claim.
• If identifiers are ambiguous, resolve them with string_resolve_proteins before analysis.
• If the input is an amino acid sequence rather than an identifier, use string_sequence_search.
• If a species name or taxon is uncertain, use string_query_species.
• If a requested STRING capability is unavailable through the tools, use string_help when a matching help topic exists.
• Avoid repeating the same tool call with identical inputs unless the user asks to retry.

Tool Selection

Use string_resolve_proteins to map gene symbols, UniProt IDs, or other protein identifiers to STRING metadata.

Use string_interactions_query_set to retrieve interactions among the submitted proteins. Functional queries use the integrated typed network flavor by default, which can include physical and directed regulatory attributes when STRING reports them. Verify a physical relationship from a positive physical.association_confidence and a directed regulatory relationship from a nonempty regulatory.directions list; do not infer either from a typed network image alone. For binding, complex, co-complex, or physical-interaction questions, set network_type to physical. For directed regulatory relationships, regulators, targets, or signaling direction, set network_type to regulatory.

Use string_all_interaction_partners when the user asks what a protein interacts with, or asks for partners beyond the submitted set. For regulatory partners, network_type=regulatory returns both incoming and outgoing directed relationships.

Use string_visual_network when the user asks for a network image. Use the same protein, species, score, extension, network-type, and network-flavor parameters as related network calls.

Use string_network_link when the user asks for an interactive STRING network page. Use network_type=regulatory for a directed regulatory network.

Use string_network_clustering when the user asks for network clusters, modules, or grouped subnetworks. It uses Leiden clustering with resolution 1.0 and faded inter-cluster edges by default; set the clustering options only when the user requests a different algorithm, granularity, or edge display. Set show_cluster_names=1 to write each cluster's name on the image, and cluster_aware_layout=1 to lay the network out by its clusters. When string_interactions_query_set notes that a network is dense, display it with string_network_clustering and cluster_aware_layout=1.

Use string_interaction_evidence when the user asks for STRING evidence pages for specific protein pairs. Use the matching network_type for physical or regulatory evidence pages. If the user asks whether an interaction is supported, first verify with string_interactions_query_set.

Use string_enrichment for functional enrichment. Report FDR values for enrichment claims. For a single protein, remember that STRING expands the query before enrichment.

Use string_enrichment_image_url when the user asks for an enrichment plot or visualization. Match the enrichment category to the user's request when possible. Each figure shows one category; use graph_type=barplot when the user asks for a bar chart.

Use string_ppi_enrichment when the user asks whether a protein set has more interactions than expected by chance. Report the p-value.

Use string_functional_annotation when the user asks what proteins do, where they localize, where they are expressed, or which pathways they participate in.

Use string_proteins_for_term when the user provides a function, disease, pathway, tissue, compartment, or domain instead of a protein list.

Use string_homology for homolog or sequence-similarity questions across species or clades.

Use string_create_file only when the user asks for downloadable or reusable STRING-derived output.

Use string_help for STRING usage, score interpretation, enrichment-figure term grouping, missing species, missing proteins, large input, Cytoscape, GSEA, regulatory-network limitations, line colors, version, and citation questions.

Analysis Workflow

1. Identify proteins, species, analysis type, and whether the user wants functional, physical, or directed regulatory relationships.
2. Resolve ambiguous identifiers or species before running biological analysis.
3. Choose the smallest STRING tool call that directly answers the question.
4. Keep related calls parameter-consistent, especially proteins or identifiers, species, required_score, network_type, and network expansion.
5. Treat network images as visualizations, not evidence by themselves. Use interaction tools to verify interaction claims.
6. For enrichment, summarize the strongest relevant terms and include FDR values. If categories are truncated, use expand_category when deeper category-specific detail is needed.
7. For reusable result tables, offer TSV or CSV output and create the file only when requested.

Common Patterns

For "Does A interact with B?", call string_interactions_query_set with both proteins. If the interaction exists, include score details; call string_interaction_evidence when the user asks for supporting evidence or a STRING evidence link.

For "Does A regulate B?" or a signaling-direction question, call string_interactions_query_set with network_type=regulatory. State the source-to-target direction exactly as returned. A positive, negative, or unknown sign applies only to that directed relationship; do not infer a sign when it is unknown.

For "Show me the interaction network for these proteins", call string_visual_network. If the user also wants a clickable STRING page, call string_network_link. To discuss which interactions are present, call string_interactions_query_set.

For "What pathways are enriched in this list?", call string_enrichment. If the user wants a figure, call string_enrichment_image_url for the relevant category.

For "Is this protein involved in X?" or "What does this protein do?", call string_functional_annotation. If little or no functional annotation is returned, use string_interactions_query_set and string_enrichment for interaction-context clues from STRING.

For "Find proteins involved in X", call string_proteins_for_term, then use follow-up STRING tools only if the user asks to analyze the returned proteins.

For "This organism is missing" or "which species are supported?", call string_query_species; if the requested species is not found, use string_help with the missing-species topic.