Skip to main content
Glama

CRC-LNM Medical Agent 1.0.14

这是面向 ModelScope 托管 STDIO 的完整懒加载纯 NumPy 单模型版本。一个 console script、一个进程内注册六个医学工具;initialize 和 tools/list 不读取模型参数或展开病例。第一次 crc_lnm_predict_multimodal 调用校验并加载唯一的 seed_2024 NumPy runtime asset,后续预测复用同一实例。默认安装不依赖 PyTorch、NVIDIA 或 CUDA 包。

更新说明 (v1.0.14)

  • 根本性重构:所有工具参数从 UUID4 / Literal / BaseModel 嵌套结构改为扁平化 str / int / float 原生类型

  • Nexent 兼容性修复:移除复杂 Pydantic 类型注解(UUID4、Literal["1.1.0"]),避免 JSON Schema 中出现 anyOf / pattern 等复杂结构导致 Nexent 解析失败

  • 临床输入扁平化:将 clinical 嵌套对象展开为 clinical_age / clinical_male / clinical_type / clinical_t 四个独立参数

  • 简化类型注解:所有工具参数现接受原生 Python 类型(str / int / float),内部转 UUID 对象

  • 稳定的 description 字段:所有 6 个工具都有非空中文 description,符合 Nexent JSON-RPC 校验

Related MCP server: DROMA MCP Server

ModelScope 正式配置

{
  "mcpServers": {
    "crc-lnm-medical-agent": {
      "command": "uvx",
      "args": [
        "crc-lnm-medical-agent@1.0.14"
      ]
    }
  }
}

ModelScope 中选择托管部署和 STDIO;command 填 uvx,args 只填上面一个带版本参数。不要增加 URL、host、port、transport 参数或环境变量。

本地验收

powershell -ExecutionPolicy Bypass -File scripts/release_verify_full.ps1

六工具为模型信息、病例质控、CT 特征准备、病理特征准备、单模型预测和报告生成。内置病例仅为合成演示资源;本包只用于科研辅助,不构成诊断,所有输出均需专家复核。

Available Tools

6 tools
crc_lnm_case_data_qcB

Validate deidentified case integrity, privacy and required modalities.

Inputs: input (required), case_ref (required), trace_id (required), request_id (required), +1 more. Returns: ``crc_lnm_case_data_qc`` 的独立响应合同。

crc_lnm_generate_reportC

Generate a deterministic escaped research-assistance report.

Inputs: input (required), case_ref (required), trace_id (required), request_id (required), +1 more. Returns: ``crc_lnm_generate_report`` 的独立响应合同。

crc_lnm_get_model_infoB
Read-onlyIdempotent

Return integrity-checked model dimensions, version, hashes and threshold.

Inputs: input (required), trace_id (required), request_id (required), contract_version (required). Returns: ``crc_lnm_get_model_info`` 的独立响应合同。

crc_lnm_predict_multimodalC

Run the locked five-member multimodal ensemble after compatibility gates.

Inputs: input (required), case_ref (required), trace_id (required), request_id (required), +1 more. Returns: ``crc_lnm_predict_multimodal`` 的独立响应合同。

crc_lnm_prepare_ct_featuresB

Validate and retain approved precomputed 1409-dimensional CT features.

Inputs: input (required), case_ref (required), trace_id (required), request_id (required), +1 more. Returns: ``crc_lnm_prepare_ct_features`` 的独立响应合同。

crc_lnm_prepare_pathology_featuresC

Validate and retain approved 768-dimensional pathology features.

Inputs: input (required), case_ref (required), trace_id (required), request_id (required), +1 more. Returns: ``crc_lnm_prepare_pathology_features`` 的独立响应合同。

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.5/5.0

    Scored across 6 tools

    Disambiguation5/5

    Each tool serves a distinct pipeline stage: model info retrieval, case data QC, CT feature preparation, pathology feature preparation, multimodal prediction, and report generation. No two tools appear to handle the same responsibility, so an agent can unambiguously select the right tool for each step.

    Naming Consistency4/5

    All tools share the consistent 'crc_lnm_' prefix and use snake_case. Most follow a verb_noun pattern (get_model_info, prepare_ct_features, generate_report), though 'case_data_qc' is more noun-like and 'predict_multimodal' uses an adjective, creating minor deviations. Overall the naming is predictable and readable.

    Tool Count5/5

    Six tools cover the full end-to-end workflow of a specialized medical AI pipeline without redundancy. The count is appropriately scoped for the server's purpose, neither sparse nor bloated.

    Completeness5/5

    The tool set covers the complete workflow from model inspection and data QC through feature preparation, prediction, and report generation. There are no obvious gaps for the intended use case, as each step in the pipeline is represented.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      C
      quality
      D
      maintenance
      A server that enables AI assistants to interact with cancer genomics data from cBioPortal, allowing users to explore cancer studies, access genomic data, and retrieve mutations and clinical information.
      17
      6
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables natural language interactions with cancer pharmacogenomics data through the DROMA platform, supporting drug-omics association analysis, dataset management, molecular profile loading, and treatment response analysis across multiple research projects.
      1
      Mozilla Public 2.0
    • F
      license
      A
      quality
      D
      maintenance
      Enables interaction with synthetic NIH-style clinical research data through tools for searching publications, querying patient metadata, analyzing AAA measurements, and retrieving protocol guidance.
      5
      -
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables searching and retrieving cancer-genomics projects, cases, and genomic data files — including TCGA, TARGET, and CPTAC — via keyless, open-access metadata tools. Supports filtering by primary site, project, or data category, and can route plain-English questions to the right tool automatically.
      90 npm
      MIT