Claude MCP connector

CyteType's MCP server lets you explore your scRNA-seq results (clusters, gene expression, annotations, differential expression) directly through Claude using natural conversation. No SDK or API keys required.

MCP server URL: https://mcp.cytetype.nygen.io

Steps

1. In Claude, click Customize in the left sidebar, then select Connectors.
2. Click + Add Custom Connector. Give it a name (for example CyteType), paste https://mcp.cytetype.nygen.io, and click Add. If it does not connect automatically, click Connect.
3. You will be taken to the CyteType login page. No existing account is needed; enter your email, verify with the one-time code sent to your inbox, and you will be redirected back to Claude.
4. Back in Claude, open the CyteType connector menu and select Always allow from the dropdown to enable all 23 tools without per-call prompts.
5. Ask Claude to connect to a job and start asking questions. Try: "Connect to my latest job and show me all clusters."

💡 Note: Public reports work without an existing account. You still log in via email OTP, but you can immediately explore any public job ID. Authenticated users also get access to their own private, shared, and organization-visible reports.

Available tools

The connector exposes 23 tools grouped by category.

Session management (4)

• connect_job — Connect by ID or interactively pick from your jobs
• list_my_jobs — Browse all jobs you own or have access to
• list_connected_jobs — See all jobs loaded in this session
• disconnect_job — Free a job's memory when you are done

Dataset exploration (8)

• list_clusters — All clusters with annotations and top markers
• get_cluster_detail — Full detail for a single cluster
• lookup_gene_expression — Mean expression of genes across clusters
• describe_metadata — Metadata schema with semantic role hints
• query_metadata — SQL queries on cell metadata via DuckDB
• run_differential_expression — DE analysis between cluster groups
• get_cell_interactions — LIANA cell-cell communication data
• cluster_metadata_enrichment — Per-cluster breakdown by metadata value

Annotation introspection (6)

• get_annotation_evidence — Gene-level evidence with literature citations
• get_annotation_context_fit — Tissue and pathway context fit
• get_cell_identity — Cell state, ontology term, granular type
• get_annotation_candidates — All candidates considered and reasoning
• get_candidate_gene_groups — Supporting and refuting gene sets
• get_embedding_coordinates — Cell coordinates for visualization

Annotation deep dive (5)

• get_quality_control — Confidence, heterogeneity, classifier accuracy
• get_expert_reviews — Expert reviewer panel assessments
• get_neighbor_degs — Differential genes vs. nearest neighbors
• get_co_expression_evidence — Co-expression pairs with statistics
• get_pathway_details — Pathway enrichment with NES scores

💡 Tip: Connect up to 10 jobs in a single session. Call connect_job with different job IDs, then pass job_id to any tool. When only one job is connected, the parameter is optional.

Example prompts

Natural language is enough; you do not need to memorize tool names.

• Connect to my latest job and show me all clusters with their annotations and confidence levels.
• What is the expression of TP53, BRCA1, and CD8A across all clusters?
• Run DE between the T cell clusters and the macrophage clusters. Show me the top upregulated genes.
• Why was cluster 2 annotated as exhausted CD8+ T cells? Show me the evidence and what else was considered.
• Which clusters have low confidence? What did the expert reviewers flag?
• Connect to both my treatment and control jobs and compare the T cell populations.

Learn more

How CyteType works

Interactive Cluster Copilot