scmcp-shared 0.5.0__tar.gz → 0.6.0__tar.gz

{scmcp_shared-0.5.0 → scmcp_shared-0.6.0}/.github/workflows/test.yml

@@ -2,13 +2,20 @@ name: test package
 
 on:
   push:
-    branches: [ "main" ]
+    branches: ["main"]
   pull_request:
-    branches: [ "main" ]
+    branches: ["main"]
 
 jobs:
   test:
     runs-on: ubuntu-latest
+    env:
+      MODEL: ${{ vars.MODEL }}
+      BASE_URL: ${{ vars.BASE_URL }}
+      API_KEY: ${{ secrets.API_KEY }}
+      EMBEDDER_MODEL: ${{ vars.EMBEDDER_MODEL }}
+      EMBEDDER_BASE_URL: ${{ vars.EMBEDDER_BASE_URL }}
+      EMBEDDER_API_KEY: ${{ secrets.EMBEDDER_API_KEY }}
     strategy:
       matrix:
         python-version: ["3.10", "3.11", "3.12", "3.13"]
@@ -31,4 +38,3 @@ jobs:
 
       - name: Run pytest
         run: uv run --no-sync pytest
-

{scmcp_shared-0.5.0 → scmcp_shared-0.6.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scmcp_shared
-Version: 0.5.0
+Version: 0.6.0
 Summary: A shared function libray for scmcphub
 Project-URL: Homepage, http://scmcphub.org/
 Project-URL: Repository, https://github.com/scmcphub/scmcp-shared
@@ -38,12 +38,15 @@ License-File: LICENSE
 Keywords: AI,agent,bioinformatics,llm,mcp,model context protocol,scRNA-seq,single cell
 Requires-Python: >=3.10
 Requires-Dist: abcoder
+Requires-Dist: agno
 Requires-Dist: fastmcp>=2.7.0
 Requires-Dist: igraph
-Requires-Dist: instructor>=1.8.3
+Requires-Dist: lancedb
 Requires-Dist: leidenalg
 Requires-Dist: mcp>=1.8.0
 Requires-Dist: nest-asyncio
+Requires-Dist: openai
+Requires-Dist: requests
 Requires-Dist: scanpy
 Description-Content-Type: text/markdown
 

{scmcp_shared-0.5.0 → scmcp_shared-0.6.0}/pyproject.toml

@@ -26,9 +26,12 @@ dependencies = [
     "leidenalg",
     "mcp>=1.8.0",
     "fastmcp>=2.7.0",
+    "openai",
+    "lancedb",
+    "agno",
     "nest_asyncio",
-    "instructor>=1.8.3",
-    "abcoder"
+    "abcoder",
+    "requests"
 ]
 
 [build-system]

scmcp_shared-0.6.0/src/scmcp_shared/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.6.0"

scmcp_shared-0.6.0/src/scmcp_shared/agent.py

@@ -0,0 +1,47 @@
+from .schema.tool import ToolList
+import os
+
+
+from agno.agent import Agent
+from agno.models.openai import OpenAILike
+from scmcp_shared.kb import load_kb
+
+model = OpenAILike(
+    id=os.getenv("MODEL"),
+    base_url=os.getenv("BASE_URL"),
+    api_key=os.getenv("API_KEY"),
+)
+
+
+def rag_agent(task, software=None):
+    knowledge_base = load_kb(software=software)
+    agent = Agent(
+        model=model,
+        knowledge=knowledge_base,
+        show_tool_calls=True,
+        search_knowledge=True,
+    )
+    query = f"""
+    <task>
+    {task}
+    </task>
+    查询知识库，给出一个用于解决任务的代码示例。返回结果格式为：
+    <code_example>
+        [code_example]
+    </code_example>
+    """
+    rep = agent.run(query)
+    return rep.content
+
+
+def select_tool(query):
+    agent = Agent(
+        model=model,
+        response_model=ToolList,
+        use_json_mode=True,
+        instructions="""
+        you are a bioinformatician, you are given a task and a list of tools, you need to select the most directly relevant tools to use to solve the task
+        """,
+    )
+    rep = agent.run(query)
+    return rep.content

scmcp_shared-0.6.0/src/scmcp_shared/kb.py

@@ -0,0 +1,139 @@
+from agno.document.chunking.agentic import AgenticChunking
+from agno.embedder.openai import OpenAIEmbedder
+from agno.models.deepseek import DeepSeek
+from agno.vectordb.lancedb import LanceDb
+from agno.knowledge.agent import AgentKnowledge
+import importlib.resources
+import os
+import requests
+import zipfile
+import tempfile
+import shutil
+from pathlib import Path
+import logging
+
+logger = logging.getLogger(__name__)
+
+embedder_id = os.getenv("EMBEDDER_MODEL")
+embedder_api_key = os.getenv("EMBEDDER_API_KEY")
+embedder_base_url = os.getenv("EMBEDDER_BASE_URL")
+model_id = os.getenv("MODEL")
+model_api_key = os.getenv("API_KEY")
+model_base_url = os.getenv("BASE_URL")
+
+# 配置信息
+config = {
+    "local_dir": "vector_db",
+    "huggingface_url": "https://huggingface.co/datasets/huangshing/scmcp_vector_db/resolve/main/vector_db.zip",
+}
+
+
+def download_vector_db(source="huggingface"):
+    """
+    下载向量数据库文件
+
+    Args:
+        source: 下载源 ("huggingface" 或 "github")
+    """
+
+    # 获取本地存储路径
+    package_path = importlib.resources.path("scmcp_shared", "")
+    local_dir = Path(package_path) / config["local_dir"]
+    local_dir.mkdir(exist_ok=True)
+
+    # 检查是否已存在
+    if (local_dir / "scmcp.lance").exists():
+        logger.info("Vector database already exists locally")
+        return str(local_dir)
+
+    logger.info(f"Downloading vector database from {source}...")
+
+    # 创建临时目录用于下载和解压
+    with tempfile.TemporaryDirectory() as temp_dir:
+        temp_path = Path(temp_dir)
+        zip_path = temp_path / "vector_db.zip"
+
+        try:
+            # 下载文件
+            if source == "huggingface":
+                url = config["huggingface_url"]
+            else:
+                raise ValueError(f"Unsupported source: {source}")
+
+            logger.info(f"Downloading from: {url}")
+            response = requests.get(url, stream=True)
+            response.raise_for_status()
+
+            with open(zip_path, "wb") as f:
+                for chunk in response.iter_content(chunk_size=8192):
+                    f.write(chunk)
+
+            # 解压文件
+            logger.info("Extracting downloaded archive...")
+            _extract_archive(zip_path, local_dir)
+
+            logger.info(f"Vector database downloaded and extracted to: {local_dir}")
+            return str(local_dir)
+
+        except requests.RequestException as e:
+            raise RuntimeError(f"Failed to download vector database: {e}")
+        except Exception as e:
+            raise RuntimeError(f"Failed to process vector database: {e}")
+
+
+def _extract_archive(archive_path, extract_dir):
+    """解压归档文件"""
+    with zipfile.ZipFile(archive_path, "r") as zip_ref:
+        zip_ref.extractall(extract_dir)
+
+    # 如果解压后只有一个子目录，移动内容到目标目录
+    extracted_items = list(Path(extract_dir).iterdir())
+    if len(extracted_items) == 1 and extracted_items[0].is_dir():
+        subdir = extracted_items[0]
+        for item in subdir.iterdir():
+            shutil.move(str(item), str(extract_dir / item.name))
+        subdir.rmdir()
+
+
+def load_kb(software=None, auto_download=True, download_source="huggingface"):
+    """
+    加载知识库
+
+    Args:
+        software: 软件名称
+        auto_download: 是否自动下载向量数据库
+        download_source: 下载源 ("huggingface" 或 "github")
+    """
+    # 获取向量数据库路径
+    try:
+        vector_db_path = importlib.resources.path("scmcp_shared", "vector_db")
+    except FileNotFoundError:
+        if auto_download:
+            logger.info("Vector database not found in package, attempting download...")
+            vector_db_path = download_vector_db(download_source)
+        else:
+            raise FileNotFoundError(
+                "Vector database not found. Set auto_download=True to download automatically, "
+                "or manually place the vector database in the scmcp_shared package."
+            )
+
+    vector_db = LanceDb(
+        table_name=software,
+        uri=vector_db_path,
+        embedder=OpenAIEmbedder(
+            id=embedder_id,
+            base_url=embedder_base_url,
+            api_key=embedder_api_key,
+        ),
+    )
+    model = DeepSeek(
+        id=model_id,
+        base_url=model_base_url,
+        api_key=model_api_key,
+    )
+    knowledge_base = AgentKnowledge(
+        chunking_strategy=AgenticChunking(model=model),
+        vector_db=vector_db,
+    )
+
+    return knowledge_base

scmcp_shared-0.6.0/src/scmcp_shared/server/__init__.py

@@ -0,0 +1,8 @@
+from __future__ import annotations
+
+from .auto import auto_mcp
+from .rag import rag_mcp
+from abcoder.server import nb_mcp
+from . import preset
+
+__all__ = ["auto_mcp", "rag_mcp", "nb_mcp", "preset"]

{scmcp_shared-0.5.0 → scmcp_shared-0.6.0}/src/scmcp_shared/server/auto.py

@@ -15,17 +15,19 @@ def search_tool(
     """search the tools and get tool parameters that can be used to solve the  user's tasks or questions"""
     ctx = get_context()
     fastmcp = ctx.fastmcp
-    all_tools = fastmcp._tool_manager._all_tools
+    if hasattr(fastmcp._tool_manager, "_all_tools"):
+        all_tools = fastmcp._tool_manager._all_tools
+    else:
+        all_tools = fastmcp._tool_manager._tools
     auto_tools = fastmcp._tool_manager._tools
     fastmcp._tool_manager._tools = all_tools
     query = f"<task>{task}</task>\n"
     for name in all_tools:
         tool = all_tools[name]
         query += f"<Tool>\n<name>{name}</name>\n<description>{tool.description}</description>\n</Tool>\n"
-
     results = select_tool(query)
     tool_list = []
-    for tool in results:
+    for tool in results.tools:
         tool = tool.model_dump()
         tool["parameters"] = all_tools[tool["name"]].parameters
         tool_list.append(tool)

scmcp_shared-0.6.0/src/scmcp_shared/server/rag.py

@@ -0,0 +1,13 @@
+from fastmcp import FastMCP
+from ..agent import rag_agent
+from pydantic import Field
+
+rag_mcp = FastMCP("RAG-Server")
+
+
+@rag_mcp.tool(tags={"rag"})
+def retrieve_knowledge(
+    task: str = Field(description="The tasks or questions that needs to be solved"),
+):
+    """search function and parameters that can be used to solve the user's tasks or questions"""
+    return rag_agent(task, software="scmcp")

{scmcp_shared-0.5.0 → scmcp_shared-0.6.0}/tests/conftest.py

@@ -6,6 +6,7 @@ from scmcp_shared.server.preset import ScanpyPreprocessingMCP
 from scmcp_shared.server.preset import ScanpyToolsMCP
 from scmcp_shared.server.preset import ScanpyPlottingMCP
 from scmcp_shared.server.preset import ScanpyUtilMCP
+from scmcp_shared.server.auto import auto_mcp
 
 
 class ScanpyMCPManager(BaseMCPManager):
@@ -19,6 +20,7 @@ class ScanpyMCPManager(BaseMCPManager):
             "tl": ScanpyToolsMCP().mcp,
             "pl": ScanpyPlottingMCP().mcp,
             "ul": ScanpyUtilMCP().mcp,
+            "auto": auto_mcp,
         }
 
 

{scmcp_shared-0.5.0 → scmcp_shared-0.6.0}/tests/test_io.py

@@ -15,9 +15,9 @@ async def test_read_and_write(mcp):
     async with Client(mcp) as client:
         # tools = await client.list_tools()
         # client._tools = tools
-        print(mcp._tool_manager._tools, 123)
         result = await client.call_tool("io_read", {"request": {"filename": test_dir}})
-        assert "AnnData" in result[0].text
+        print(result)
+        assert "AnnData" in result.content[0].text
 
         result = await client.call_tool("io_write", {"request": {"filename": outfile}})
         assert outfile.exists()

{scmcp_shared-0.5.0 → scmcp_shared-0.6.0}/tests/test_pp.py

@@ -13,17 +13,17 @@ async def test_complete_workflow(mcp):
     async with Client(mcp) as client:
         # Load the data
         result = await client.call_tool("io_read", {"request": {"filename": test_dir}})
-        assert "AnnData" in result[0].text
+        assert "AnnData" in result.content[0].text
 
         # Filter cells and genes
         result = await client.call_tool(
             "pp_subset_cells", {"request": {"min_genes": 200}}
         )
-        assert "AnnData" in result[0].text
+        assert "AnnData" in result.content[0].text
         result = await client.call_tool(
             "pp_subset_genes", {"request": {"min_cells": 3}}
         )
-        assert "AnnData" in result[0].text
+        assert "AnnData" in result.content[0].text
 
         # Add mitochondrial genes and calculate QC metrics
         result = await client.call_tool(
@@ -36,73 +36,73 @@ async def test_complete_workflow(mcp):
                 }
             },
         )
-        assert "mt" in result[0].text
+        assert "mt" in result.content[0].text
         result = await client.call_tool(
             "pp_calculate_qc_metrics",
             {"request": {"qc_vars": ["mt"], "percent_top": None, "log1p": False}},
         )
-        assert "total_counts" in result[0].text
+        assert "total_counts" in result.content[0].text
 
         # Filter based on QC metrics
         result = await client.call_tool(
             "pp_subset_cells", {"request": {"max_genes": 2500, "max_mt_percent": 5}}
         )
-        assert "AnnData" in result[0].text
+        assert "AnnData" in result.content[0].text
 
         # Normalize and transform
         result = await client.call_tool(
             "pp_normalize_total", {"request": {"target_sum": 1e4}}
         )
-        assert "AnnData" in result[0].text
+        assert "AnnData" in result.content[0].text
         result = await client.call_tool("pp_log1p", {"request": {}})
-        assert "log1p" in result[0].text
+        assert "log1p" in result.content[0].text
 
         # Find highly variable genes
         result = await client.call_tool(
             "pp_highly_variable_genes",
             {"request": {"min_mean": 0.0125, "max_mean": 3, "min_disp": 0.5}},
         )
-        assert "highly_variable" in result[0].text
+        assert "highly_variable" in result.content[0].text
 
         # Subset to highly variable genes
         result = await client.call_tool(
             "pp_subset_genes", {"request": {"highly_variable": True}}
         )
-        assert "AnnData" in result[0].text
+        assert "AnnData" in result.content[0].text
 
         # Regress out effects
         result = await client.call_tool(
             "pp_regress_out", {"request": {"keys": ["total_counts", "pct_counts_mt"]}}
         )
-        assert "AnnData" in result[0].text
+        assert "AnnData" in result.content[0].text
 
         # Scale the data
         result = await client.call_tool("pp_scale", {"request": {"max_value": 10}})
-        assert "AnnData" in result[0].text
+        assert "AnnData" in result.content[0].text
 
         # PCA
         result = await client.call_tool(
             "tl_pca", {"request": {"n_comps": 50, "svd_solver": "arpack"}}
         )
-        assert "X_pca" in result[0].text
+        assert "X_pca" in result.content[0].text
 
         # Compute neighborhood graph
         result = await client.call_tool(
             "pp_neighbors", {"request": {"n_neighbors": 10, "n_pcs": 40}}
         )
-        assert "neighbors" in result[0].text
+        assert "neighbors" in result.content[0].text
 
         # Leiden clustering
         result = await client.call_tool("tl_leiden", {"request": {}})
-        assert "leiden" in result[0].text
+        assert "leiden" in result.content[0].text
 
         # UMAP
         result = await client.call_tool("tl_umap", {"request": {}})
-        assert "X_umap" in result[0].text
+        assert "X_umap" in result.content[0].text
 
         # Plot UMAP
         result = await client.call_tool(
             "pl_embedding",
             {"request": {"basis": "umap", "color": ["leiden", "NKG7", "PPBP"]}},
         )
-        assert "figure" in result[0].text
+        assert "figure" in result.content[0].text

scmcp_shared-0.6.0/tests/test_select_tool.py

@@ -0,0 +1,16 @@
+import pytest
+from fastmcp import Client
+import nest_asyncio
+
+# Apply nest_asyncio at module level
+nest_asyncio.apply()
+
+
+@pytest.mark.asyncio
+async def test_tool_selection(mcp):
+    task = """
+    read the file  /date/test.h5ad , filter the cells with ncells>100
+    """
+    async with Client(mcp) as client:
+        result = await client.call_tool("auto_search_tool", {"task": task})
+        assert "io_read" in result.content[0].text

scmcp_shared-0.5.0/src/scmcp_shared/__init__.py

@@ -1 +0,0 @@
-__version__ = "0.5.0"

scmcp_shared-0.5.0/src/scmcp_shared/agent.py

@@ -1,80 +0,0 @@
-import instructor
-from openai import OpenAI
-from scmcp_shared.schema.tool import ToolList
-import os
-
-
-from agno.agent import Agent
-from agno.models.openai import OpenAILike
-from scmcp_shared.kb import load_kb
-
-model = OpenAILike(
-    id=os.getenv("MODEL"),
-    base_url=os.getenv("BASE_URL"),
-    api_key=os.getenv("API_KEY"),
-)
-
-
-def rag_agent(task, software=None):
-    knowledge_base = load_kb(software=software)
-    agent = Agent(
-        model=model,
-        knowledge=knowledge_base,
-        show_tool_calls=True,
-        search_knowledge=True,
-    )
-    query = f"""
-    <task>
-    {task}
-    </task>
-    拆分任务，分别查询知识库，查找完成每个子任务所需要的函数以及参数, 给出一个总的用于解决任务的代码示例。返回结果格式为：
-    <function_list>
-        <function>
-            <package>.<module>.<function_name>
-            <package>.<module>.<function_name>(<param1>[, <param2>=<default>][, ...][, **kwargs])
-            <一句话描述函数的主要功能>。
-
-            <可选：补充说明/实现细节/依赖等>
-
-            Parameters
-            :
-            <param1> (<type>) – <param1 的说明>
-
-            <param2> (<type> (default: <default>)) – <param2 的说明>
-
-            <param3> (<type> (default: <default>)) – <param3 的说明>
-
-            ...（按需继续列出其他参数）
-
-            **kwargs – <对底层实现传递的其他关键字参数的说明；如无需说明可简写为“其他关键字参数”。>
-        </function>
-                        ...
-    </function_list>
-    <code_example>
-    [code_example]
-    </code_example>
-    """
-    rep = agent.run(query)
-    return rep.content
-
-
-def select_tool(query):
-    API_KEY = os.environ.get("API_KEY", None)
-    BASE_URL = os.environ.get("BASE_URL", None)
-    MODEL = os.environ.get("MODEL", None)
-
-    client = OpenAI(api_key=API_KEY, base_url=BASE_URL)
-    client = instructor.from_openai(client)
-
-    response = client.chat.completions.create(
-        model=MODEL,
-        messages=[
-            {
-                "role": "system",
-                "content": "you are a bioinformatician, you are given a task and a list of tools, you need to select the most directly relevant tools to use to solve the task",
-            },
-            {"role": "user", "content": query},
-        ],
-        response_model=ToolList,
-    )
-    return response.tools

scmcp_shared-0.5.0/src/scmcp_shared/kb.py

@@ -1,38 +0,0 @@
-from agno.document.chunking.agentic import AgenticChunking
-from agno.embedder.openai import OpenAIEmbedder
-from agno.models.deepseek import DeepSeek
-from agno.vectordb.lancedb import LanceDb
-from agno.knowledge.agent import AgentKnowledge
-import importlib.resources
-import os
-
-
-embedder_id = os.getenv("EMBEDDER_MODEL")
-embedder_api_key = os.getenv("EMBEDDER_API_KEY")
-embedder_base_url = os.getenv("EMBEDDER_BASE_URL")
-model_id = os.getenv("MODEL")
-model_api_key = os.getenv("API_KEY")
-model_base_url = os.getenv("BASE_URL")
-
-
-def load_kb(software=None):
-    vector_db = LanceDb(
-        table_name=software,
-        uri=importlib.resources.path("scmcp_shared", "vector_db"),
-        embedder=OpenAIEmbedder(
-            id=embedder_id,
-            base_url=embedder_base_url,
-            api_key=embedder_api_key,
-        ),
-    )
-    model = DeepSeek(
-        id=model_id,
-        base_url=model_base_url,
-        api_key=model_api_key,
-    )
-    knowledge_base = AgentKnowledge(
-        chunking_strategy=AgenticChunking(model=model),
-        vector_db=vector_db,
-    )
-
-    return knowledge_base

scmcp_shared-0.5.0/src/scmcp_shared/server/__init__.py

@@ -1 +0,0 @@
-from __future__ import annotations

scmcp_shared-0.5.0/src/scmcp_shared/vector_db/decoupler.lance/_transactions/0-9499b1b5-85d4-44c2-8f05-1bcce87fe4ef.txn

@@ -1,3 +0,0 @@
-$9499b1b5-85d4-44c2-8f05-1bcce87fe4ef��
<vector ���������
*fixed_size_list:float:10240
8
Zdefault&id
 ���������
*string0
8Zdefault+payload ���������
*string0
8Zdefault"!
-lance.auto_cleanup.interval20"'
-lance.auto_cleanup.older_than14days