SCMCP

Instructions

Implementation Reference

• src/scmcp/tool/io.py:28-44 (handler)

  Core handler function implementing the reading logic for various file formats using scanpy.read or sc.read_10x_mtx.
  def read_func(**kwargs): file = Path(kwargs["filename"]) if file.is_dir(): kwargs["path"] = kwargs["filename"] parameters = inspect.signature(sc.read_10x_mtx).parameters func_kwargs = {k: kwargs.get(k) for k in parameters if k in kwargs} adata = sc.read_10x_mtx(**func_kwargs) elif file.is_file(): parameters = inspect.signature(sc.read).parameters func_kwargs = {k: kwargs.get(k) for k in parameters if k in kwargs} logger.info(func_kwargs) adata = sc.read(**func_kwargs) if not kwargs.get("first_column_obs", True): adata = adata.T else: adata = "there are no file" return adata
• src/scmcp/tool/io.py:70-84 (handler)

  Dispatch handler for IO tools, specifically handles read_tool by managing AnnData state and calling read_func.
  def run_io_func(ads, func, arguments): if func == "read_tool": adata_id = f"adata{len(ads.adata_dic)}" if arguments.get("sampleid", None) is not None: adata_id = arguments["sampleid"] else: adata_id = f"adata{len(ads.adata_dic)}" res = read_func(**arguments) ads.active = adata_id ads.adata_dic[adata_id] = res return res else: adata = ads.adata_dic[ads.active] return write_func(adata, func, arguments)
• src/scmcp/schema/io.py:13-92 (schema)

  Pydantic model defining the input schema for the read_tool, including all parameters and validators.
  class ReadModel(JSONParsingModel): """Input schema for the read tool.""" filename: str = Field( description="Path to the file to read." ) sampleid: Optional[str] = Field( default=None, description="Sample identifier to mark and distinguish different samples." ) backed: Optional[Literal['r', 'r+']] = Field( default=None, description="If 'r', load AnnData in 'backed' mode instead of fully loading it into memory ('memory' mode). If you want to modify backed attributes of the AnnData object, you need to choose 'r+'." ) sheet: Optional[str] = Field( default=None, description="Name of sheet/table in hdf5 or Excel file." ) ext: Optional[str] = Field( default=None, description="Extension that indicates the file type. If None, uses extension of filename." ) delimiter: Optional[str] = Field( default=None, description="Delimiter that separates data within text file. If None, will split at arbitrary number of white spaces, which is different from enforcing splitting at any single white space." ) first_column_names: bool = Field( default=False, description="Assume the first column stores row names. This is only necessary if these are not strings: strings in the first column are automatically assumed to be row names." ) first_column_obs: bool = Field( default=True, description="If True, assume the first column stores observations (cell or barcode) names when provide text file. If False, the data will be transposed." ) backup_url: Optional[str] = Field( default=None, description="Retrieve the file from an URL if not present on disk." ) cache: bool = Field( default=False, description="If False, read from source, if True, read from fast 'h5ad' cache." ) cache_compression: Optional[Literal['gzip', 'lzf']] = Field( default=None, description="See the h5py dataset_compression. (Default: settings.cache_compression)" ) var_names: Optional[str] = Field( default="gene_symbols", description="The variables index for 10x mtx format. Either 'gene_symbols' or 'gene_ids'." ) make_unique: bool = Field( default=True, description="Whether to make the variables index unique by appending '-1', '-2' etc. or not. Used for 10x mtx format." ) gex_only: bool = Field( default=True, description="Only keep 'Gene Expression' data and ignore other feature types, e.g. 'Antibody Capture', 'CRISPR Guide Capture', or 'Custom'. Used for 10x formats." ) prefix: Optional[str] = Field( default=None, description="Any prefix before matrix.mtx, genes.tsv and barcodes.tsv. For instance, if the files are named patientA_matrix.mtx, patientA_genes.tsv and patientA_barcodes.tsv the prefix is patientA_. Used for 10x mtx format." ) @field_validator('backed') def validate_backed(cls, v: Optional[str]) -> Optional[str]: if v is not None and v not in ['r', 'r+']: raise ValueError("If backed is provided, it must be either 'r' or 'r+'") return v @field_validator('cache_compression') def validate_cache_compression(cls, v: Optional[str]) -> Optional[str]: if v is not None and v not in ['gzip', 'lzf']: raise ValueError("cache_compression must be either 'gzip', 'lzf', or None") return v @field_validator('var_names') def validate_var_names(cls, v: Optional[str]) -> Optional[str]: if v is not None and v not in ['gene_symbols', 'gene_ids']: raise ValueError("var_names must be either 'gene_symbols' or 'gene_ids'") return v
• src/scmcp/tool/io.py:15-19 (registration)

  Definition of the MCP Tool object for read_tool, specifying name, description, and input schema.
  read_tool = types.Tool( name="read_tool", description="Read data from various file formats (h5ad, 10x, text files, etc.) or directory path.", inputSchema=ReadModel.model_json_schema(), )
• src/scmcp/tool/io.py:52-55 (registration)

  Registration of the read_tool in the io_tools dictionary, which is used by the MCP server for listing and dispatching tools.
  io_tools = { "read_tool": read_tool, "write_tool": write_tool, }