Loading ToolBench

Reading the current corpus…

scmcp MCP server · F · ToolBench

← Browse

scmcp

github.com/scmcphub/scmcp

An MCP server for scRNA-Seq analysis with natural language!

2026-07-28+STDIO12 tools

30/100

Definition Quality17

Protocol Readiness35

Supportability49

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

This MCP server delegates tool definitions to upstream module managers (scanpy-mcp, cellrank-mcp, infercnv-mcp, liana-mcp, decoupler-mcp) via composition. The source code in src/scmcp/server/__init__.py shows only tool SELECTION and MODULE AGGREGATION, no actual tool schema definitions, parameter validation, descriptions, or error handling are visible in the provided codebase. Tool descriptions like 'Preprocessing tool from cellrank-mcp module' and 'Visualization tool from cellrank-mcp module' are generic placeholders (12-50 chars) with no parameter guidance. Without access to the upstream module implementations, the actual schemas cannot be verified. The aggregation pattern itself is reasonable for composability, but the lack of schema visibility, parameter documentation, and error handling guidance at this server level is a critical gap.

Tools (12)

ccc_dotplotread onlysource verified22/100

Cell-cell communication dot plot visualization from liana-mcp module

chromosome_heatmapread only22/100

Visualization tool from infercnv-mcp module

circle_plotread only22/100

Circle plot visualization from liana-mcp module

circular_projectionread only22/100

Visualization tool from cellrank-mcp module

cnv_scoreread only22/100

Copy number variation scoring tool from infercnv-mcp module

communicateread onlysource verified18/100

Cell-cell communication analysis tool from liana-mcp module

filter_and_normalizewrite20/100

Preprocessing tool from cellrank-mcp module

infercnvread onlysource verified17/100

Copy number variation analysis tool from infercnv-mcp module

Issues (6)

CRITICALtool, constrained-input ↗

No input schemas visible for any of the 12 tools. All tools are delegated to upstream module managers, but no JSON Schema definitions, parameter types, constraints, or descriptions are provided in the scmcp server codebase itself. Cannot verify parameter typing, required/optional status, enums, or defaults.

filter_and_normalizekernel_projectioncircular_projectionchromosome_heatmapinfercnvcnv_scoreload_gene_positioncommunicaterank_aggregatels_ccc_methodccc_dotplotcircle_plot

CRITICALtool-description ↗

Tool descriptions are generic module-source placeholders (10-50 chars) with no actionable guidance. E.g. 'Preprocessing tool from cellrank-mcp module' does not explain what data transformation happens, when to use it, expected inputs, or outputs. Descriptions must be 50-200 chars and answer WHAT, WHEN, and WHAT-RETURNS.

filter_and_normalizekernel_projectioncircular_projectionchromosome_heatmapinfercnv

Recommendations

Override or enhance tool descriptions in the SCMCPManager or configuration layer. Do not rely on upstream module descriptions, they are often generic. Write domain-specific descriptions for each tool: e.g. 'filter_and_normalize: Filter cells by QC metrics (nGenes, percent_mt) and normalize library sizes. Preprocesses raw count matrix before downstream analysis.'
Document parameter schemas for each tool at the server level. Create a tools_schema.json or include parameter docs in the tool registration. Specify: parameter names, types (string, number, array, enum), whether they are required, constraints (min/max, pattern, choices), and examples.
Add output schema documentation. For visualization tools (kernel_projection, ccc_dotplot), document that they return file paths. For analysis tools (communicate, infercnv), document return structure (dataframe columns, statistics, p-values, metadata).
Rename tools to use clear action verbs: 'infercnv' → 'run_infercnv' or 'analyze_cnv'; 'communicate' → 'analyze_ccc' or 'predict_cell_communication'; 'rank_aggregate' → 'aggregate_ranks'; 'circle_plot' → 'generate_circle_plot'; 'ccc_dotplot' → 'generate_ccc_dotplot'.
Add error handling guidance in tool descriptions. E.g. 'If analysis fails with sparse data, try filter_and_normalize first.' or 'Returns error if gene not found in reference, use ls_genes to list available genes.'
Document data dependencies. Tools likely require AnnData objects with specific slots (X, obsm, obsp, var). State: 'Requires adata with obsm["X_umap"] computed. Call kernel_projection first if not present.'

Score history

Overall score trend

↑ 30 points across a rubric change (v1 → v2)

30/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-22	F	30	2026-07-28+	v2
2026-03-09	F	0	-	v1