@hupan56/wlkj 2.7.11 → 3.0.0

package/templates/qoder/scripts/{kg_mcp_server.py → mcp/kg_mcp_server.py}

@@ -1,5 +1,11 @@
 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
+# v3.0 路径自举: 引导到 common/bootstrap, 统一 sys.path 逻辑
+import os as _o, sys as _s
+_cp = _o.path.join(_o.path.dirname(_o.path.dirname(_o.path.abspath(__file__))), 'common')
+if _cp not in _s.path: _s.path.insert(0, _cp)
+from bootstrap import setup; setup()
+
 """
 Knowledge Graph MCP Server for QoderWork (pure Python, no SDK dependency).
 
@@ -27,18 +33,22 @@ except Exception:
     pass
 
 # Locate the scripts dir
-SCRIPTS_DIR = os.path.dirname(os.path.abspath(__file__))
-BASE_DIR = os.path.dirname(os.path.dirname(SCRIPTS_DIR))  # repo root (.qoder 的上级)
+SCRIPTS_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))  # 子包→scripts/根
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.dirname(SCRIPTS_DIR)))  # 多一层: 子包→scripts→repo  # repo root (.qoder 的上级)
 sys.path.insert(0, SCRIPTS_DIR)
+_KG_DIR = os.path.dirname(os.path.abspath(__file__))
+if _KG_DIR not in sys.path: sys.path.insert(0, _KG_DIR)  # 同包import
 
-# Suppress all non-JSON output on stdout (MCP protocol uses stdout exclusively)
-# Redirect any stray prints to stderr
-_real_stdout = sys.stdout
-sys.stdout = sys.stderr  # default: prints go to stderr
+# stdout 重定向 + 协议层由 BaseMCPServer 接管 (v3.0 重构)
+# _real_stdout 不再模块级定义, 改为运行时从基类取 (见 _capture)
 
 
 def _capture(func, *args, **kwargs):
-    """Run a function, capture its stdout (which the scripts print to)."""
+    """Run a function, capture its stdout (which the scripts print to).
+
+    v3.0: sys.stdout 在 run() 后已是 stderr (基类 _setup_streams 重定向),
+    _real_stdout 存在 BaseMCPServer._real_stdout。这里捕获时临时换回 buf。
+    """
     old = sys.stdout
     buf = io.StringIO()
     sys.stdout = buf
@@ -246,34 +256,38 @@ TOOLS = [
 
 
 # Role-based section filtering for context_pack output
-ROLE_SECTIONS = {
-    'pm':     {'1', '2', '2b', '3', '4', '5', '7'},  # PM: no code files, no style refs
-    'design': {'2', '2b', '3', '6'},                  # Design: pages+layout+fields+style
-    'dev':    {'1', '3', '5', '7'},                   # Dev: code+fields+API+wiki
-    'test':   {'1', '4', '5'},                        # Test: code+PRD+API
-    'admin':  {'4', '5'},                             # Admin: PRD+API overview
-}
-
-def _filter_by_role(text, role):
-    """Filter context_pack output by role: keep only relevant sections."""
+# v3.0: 单一真源移到 common/roles.py, 消除与 kg.py 的逐字重复
+try:
+    from common.roles import ROLE_SECTIONS, filter_by_role as _filter_by_role
+except Exception:
+    # 回退: 内联定义 (common 不可用时)
     import re
-    visible = ROLE_SECTIONS.get(role.lower())
-    if not visible:
-        return text
-    sections = re.split(r'(\n## \d+[ab]?\.)', text)
-    result = sections[0]  # header before first section
-    i = 1
-    while i < len(sections):
-        header = sections[i]
-        num_m = re.match(r'\n## (\d+[ab]?)\.', header)
-        if num_m and num_m.group(1) in visible:
-            result += header
-            if i + 1 < len(sections):
-                result += sections[i + 1]
-            i += 2
-        else:
-            i += 2  # skip this section
-    return result.strip()
+    ROLE_SECTIONS = {
+        'pm':     {'1', '2', '2b', '3', '4', '5', '7'},
+        'design': {'2', '2b', '3', '6'},
+        'dev':    {'1', '3', '5', '7'},
+        'test':   {'1', '4', '5'},
+        'admin':  {'4', '5'},
+    }
+
+    def _filter_by_role(text, role):
+        visible = ROLE_SECTIONS.get((role or '').lower())
+        if not visible:
+            return text
+        sections = re.split(r'(\n## \d+[ab]?\.)', text)
+        result = sections[0]
+        i = 1
+        while i < len(sections):
+            header = sections[i]
+            num_m = re.match(r'\n## (\d+[ab]?)\.', header)
+            if num_m and num_m.group(1) in visible:
+                result += header
+                if i + 1 < len(sections):
+                    result += sections[i + 1]
+                i += 2
+            else:
+                i += 2
+        return result.strip()
 
 
 # ── Tool implementations (added 2026-06-17) ──────────────────────────
@@ -281,107 +295,27 @@ def _filter_by_role(text, role):
 # search_wiki 走 wiki-index.json(730词)。均不依赖已失效的旧 JSON 索引。
 
 def _tool_get_workflow(module):
-    """业务操作链查询。直查 DuckDB workflows 表, 带脏数据清洗 + 模糊匹配。
-
-    审核: workflows 表 168 行/13 模块, 脏数据约 25 行(endpoint='None' 或
-    button 泄漏模板语法 '{{'), 清洗后链路完整。模块名均为英文, 中文输入会
-    模糊查 entities.cn 返回候选。
-    """
-    import kg_duckdb as K
+    """业务操作链查询 — v3.0 f3: 转发到 common.kg_capabilities (与 kg.py 共用单一真源)。"""
     q = (module or '').strip()
     if not q:
         return "请提供模块名。可选: assets/cost/hr/maintenance/operation/qualityAndSafety/veh/settings/dashboard/flow/systems/monitor/quasafe"
+    import kg_duckdb as K
+    from common.kg_capabilities import workflow_query
     con = K.get_db()
     try:
-        # 1) 精确 + 模糊匹配 workflows.module (英文)
-        hits = con.execute(
-            "SELECT DISTINCT module FROM workflows WHERE module = ? OR module LIKE '%' || ? || '%' ORDER BY module",
-            [q.lower(), q.lower()]
-        ).fetchall()
-        if not hits:
-            # 2) 中文兜底: 查 entities 里 cn 匹配的, 给候选
-            cn_hits = con.execute(
-                "SELECT canonical, cn FROM entities WHERE type IN ('FEATURE','MODULE') AND cn <> '' AND (cn LIKE '%' || ? OR cn LIKE '%' || ? || '%')",
-                [q, q]
-            ).fetchall()
-            if cn_hits:
-                cands = ', '.join('%s(%s)' % (c[0], c[1]) for c in cn_hits[:8])
-                return "未在 workflows 表找到模块 '%s'。但 entities 表有相近的中文名, 可换这些英文模块名重试:\n  %s" % (q, cands)
-            all_mods = [r[0] for r in con.execute("SELECT DISTINCT module FROM workflows ORDER BY module").fetchall()]
-            return "未找到模块 '%s'。现有模块: %s" % (q, ', '.join(all_mods))
-
-        out = []
-        for (mod,) in hits:
-            # 清洗: 过滤空端点和模板泄漏按钮
-            rows = con.execute("""
-                SELECT state, seq, button, endpoint FROM workflows
-                WHERE module = ?
-                  AND endpoint <> 'None' AND endpoint <> ''
-                  AND button NOT LIKE '{{%'
-                ORDER BY seq, state
-            """, [mod]).fetchall()
-            if not rows:
-                continue
-            # 按 seq 分组展示操作链
-            out.append('=== %s (%d 步, 已清洗脏数据) ===' % (mod, len(rows)))
-            cur_seq = -1
-            for st, sq, btn, ep in rows:
-                if sq != cur_seq:
-                    out.append('  [%d] %s' % (sq, st))
-                    cur_seq = sq
-                out.append('        %-16s -> %s' % (btn, ep))
-            out.append('')
-        return '\n'.join(out).strip() or "模块 '%s' 清洗后无有效操作链" % q
+        return workflow_query(con, q)
     finally:
         try: con.close()
         except Exception: pass
 
 
 def _tool_search_wiki(keyword):
-    """Repo Wiki 搜索。只读 wiki-index.json, 不触发自动构建(慢操作, 不适合 MCP 同步调用)。
-
-    审核: repowiki.search_wiki 函数在, wiki-index.json 有 730 关键词。
-    search_index.search_wiki_cli 会在索引缺失时调 build_wiki_index (可能跑很久),
-    这里绕开它直接调底层 search_wiki, 索引缺失即报错让用户手动跑。
-    """
+    """Repo Wiki 搜索 — v3.0 f3: 转发到 common.kg_capabilities.wiki_query (CLI/MCP共用)。"""
     kw = (keyword or '').strip()
     if not kw:
         return "请提供搜索关键词"
-    from common.paths import DATA_INDEX_DIR as _idx, PROJECT_ROOT as _base
-    wiki_path = os.path.join(str(_idx), 'wiki-index.json')
-    if not os.path.isfile(wiki_path):
-        return ("wiki-index.json 不存在。请在终端手动构建(一次性):\n"
-                "  python .qoder/scripts/search_index.py wiki\n"
-                "(需要先在 Qoder IDE 对项目生成 Repo Wiki)")
-    try:
-        from common.repowiki import search_wiki
-        from common.terms import get_cn_map_with_auto
-    except ImportError as e:
-        return "Wiki 搜索模块缺失: %s" % str(e)
-    code_dir = os.path.join(_base, 'data', 'code')
-    try:
-        cn_map = get_cn_map_with_auto()
-        results = search_wiki(kw, str(_idx), code_dir, cn_map)
-    except Exception as e:
-        return "Wiki 搜索出错: %s" % str(e)
-    if not results:
-        return "No wiki pages match: %s" % kw
-    out = ['Wiki pages matching: %s (%d found)' % (kw, len(results)), '']
-    for i, r in enumerate(results, 1):
-        out.append('  %d. [%s] %s' % (i, r.get('project', '?'), r.get('title', '')))
-        deps = r.get('dependent_files') or []
-        if deps:
-            out.append('     依赖: %s%s' % (', '.join(deps[:2]),
-                      (' (共 %d 个文件)' % len(deps)) if len(deps) > 2 else ''))
-        if r.get('md_path'):
-            try:
-                rel = os.path.relpath(r['md_path'], _base)
-            except ValueError:
-                rel = r['md_path']
-            out.append('     文档: %s' % rel)
-    out.append('')
-    out.append('  提示: 读 wiki 文档 (带行号引用) 比读源码更省 token')
-    return '\n'.join(out)
+    from common.kg_capabilities import wiki_query
+    return wiki_query(kw)
 
 
 def _tool_multi_hop(symbol, max_depth=3):
@@ -402,70 +336,9 @@ def _tool_multi_hop(symbol, max_depth=3):
     depth = max(1, min(5, depth))  # 钳制 1-5
     con = K.get_db()
     try:
-        # 1) 尝试解析为 entity_id: 直接匹配 / 别名表 / 模糊匹配 entity_id
-        start_id = sym
-        # 1a) 别名精确匹配 (小写)
-        resolved = con.execute(
-            "SELECT entity_id FROM aliases WHERE alias = ?", [sym.lower()]
-        ).fetchone()
-        if resolved:
-            start_id = resolved[0]
-        else:
-            # 1b) entity_id 模糊匹配 (用户传端点 /asset/list 或函数名)
-            fuzzy = con.execute(
-                "SELECT id FROM entities WHERE id = ? OR id LIKE '%' || ? || '%' LIMIT 1",
-                [sym, sym]
-            ).fetchone()
-            if fuzzy:
-                start_id = fuzzy[0]
-            else:
-                # 1c) edges 里直接含该符号作为端点
-                edge_hit = con.execute(
-                    "SELECT from_id FROM edges WHERE from_id LIKE '%' || ? || '%' LIMIT 1",
-                    [sym]
-                ).fetchone()
-                if edge_hit:
-                    start_id = edge_hit[0]
-
-        # 2) 递归 CTE 遍历
-        try:
-            hops = K.multi_hop(con, start_id, max_depth=depth)
-        except Exception as e:
-            return "遍历出错 (起点=%s): %s" % (start_id, str(e))
-
-        if not hops or (len(hops) == 1 and hops[0][1] == 0):
-            return "从 '%s' (解析为 %s) 出发无可达节点。可能该符号不在图谱中, 或无出边。" % (sym, start_id)
-
-        # 3) 按 depth 分组, 取每层的实体详情 (最多展示)
-        from collections import defaultdict
-        by_depth = defaultdict(list)
-        for eid, d in hops:
-            by_depth[d].append(eid)
-
-        out = ['=== 多跳遍历: %s (解析为 %s) ===' % (sym, start_id)]
-        out.append('可达 %d 个实体, 最大 %d 跳:' % (len(hops), max(by_depth.keys())))
-        out.append('')
-        for d in sorted(by_depth.keys()):
-            eids = by_depth[d]
-            out.append('[第 %d 跳] %d 个实体:' % (d, len(eids)))
-            # 拉每个实体的 canonical/cn 信息
-            shown = 0
-            for eid in eids:
-                if shown >= 12:
-                    out.append('  ... 还有 %d 个' % (len(eids) - 12))
-                    break
-                info = con.execute(
-                    "SELECT type, canonical, cn FROM entities WHERE id = ?", [eid]
-                ).fetchone()
-                if info:
-                    t, c, cn = info
-                    label = '%s / %s' % (c, cn) if cn and cn != c else c
-                    out.append('  %-14s %-12s %s' % ('[%s]' % t, eid[:40], label))
-                else:
-                    out.append('  %-14s %s' % ('[edge]', eid[:60]))
-                shown += 1
-            out.append('')
-        return '\n'.join(out).strip()
+        # v3.0 f3: 核心查询转发到 common.kg_capabilities.hop_query (CLI/MCP共用)
+        from common.kg_capabilities import hop_query
+        return hop_query(con, sym, depth, multi_hop_fn=K.multi_hop)
     finally:
         try: con.close()
         except Exception: pass
@@ -703,99 +576,31 @@ def _execute_tool(name, arguments):
     return "未知工具: %s" % name
 
 
-# ── MCP Stdio Protocol (JSON-RPC 2.0) ─────────────────────────────────
-
-def _send(msg):
-    """Send a JSON-RPC message on stdout (MCP transport)."""
-    _real_stdout.write(json.dumps(msg, ensure_ascii=False) + '\n')
-    _real_stdout.flush()
-
-
-async def _handle(req):
-    """Handle a single JSON-RPC request."""
-    method = req.get("method", "")
-    req_id = req.get("id")
-    params = req.get("params", {})
-
-    if method == "initialize":
-        _send({
-            "jsonrpc": "2.0", "id": req_id,
-            "result": {
-                "protocolVersion": "2024-11-05",
-                "capabilities": {"tools": {}},
-                "serverInfo": {"name": "qoder-knowledge-graph", "version": "1.0.0"},
-            }
-        })
-
-    elif method == "notifications/initialized":
-        pass  # no response needed for notifications
-
-    elif method == "tools/list":
-        _send({"jsonrpc": "2.0", "id": req_id, "result": {"tools": TOOLS}})
-
-    elif method == "tools/call":
-        tool_name = params.get("name", "")
-        arguments = params.get("arguments", {})
-        try:
-            result_text = await asyncio.get_event_loop().run_in_executor(
-                None, _execute_tool, tool_name, arguments)
-            _send({
-                "jsonrpc": "2.0", "id": req_id,
-                "result": {
-                    "content": [{"type": "text", "text": result_text}],
-                    "isError": False,
-                }
-            })
-        except Exception as e:
-            _send({
-                "jsonrpc": "2.0", "id": req_id,
-                "result": {
-                    "content": [{"type": "text", "text": "[error] %s" % str(e)}],
-                    "isError": True,
-                }
-            })
+# ── MCP Server (继承 BaseMCPServer, 协议层由基类接管) ──────────────────
+# v3.0 重构: 删除 _send/_handle/main 三段逐字复制代码 (~96行),
+# 改为继承 BaseMCPServer。kg 的 initialize 不做内网校验 (本地知识图谱)。
 
-    elif method == "ping":
-        _send({"jsonrpc": "2.0", "id": req_id, "result": {}})
+def _bootstrap_base():
+    """延迟 import BaseMCPServer (scripts/ 在 sys.path)。"""
+    sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+    from common.mcp_base import BaseMCPServer
+    return BaseMCPServer
 
-    else:
-        if req_id is not None:
-            _send({"jsonrpc": "2.0", "id": req_id,
-                   "error": {"code": -32601, "message": "Method not found: %s" % method}})
 
+class KnowledgeGraphMCPServer(_bootstrap_base()):
+    """知识图谱 MCP Server。继承基类协议层, 只填工具逻辑。
 
-def main():
-    """Read JSON-RPC requests from stdin (synchronous readline), dispatch.
-
-    Synchronous is more reliable than asyncio on Windows Proactor event loop.
-    Tool execution runs in a thread pool to avoid blocking on long queries.
+    与 mysql/zentao 不同: kg 不做内网校验 (本地 DuckDB),
+    所以不覆写 on_initialize_hook (基类默认空)。
     """
-    import threading
 
-    while True:
-        try:
-            line = sys.stdin.buffer.readline()
-        except (EOFError, OSError):
-            break
-        if not line:
-            break
-        line = line.strip()
-        if not line:
-            continue
-        try:
-            req = json.loads(line.decode('utf-8'))
-            # Handle synchronously (tools are fast enough; context_pack ~1-2s)
-            loop = asyncio.new_event_loop()
-            loop.run_until_complete(_handle(req))
-            loop.close()
-        except json.JSONDecodeError:
-            pass
-        except Exception as e:
-            print('[kg-mcp] error: %s' % str(e), file=sys.stderr)
+    SERVER_NAME = "qoder-knowledge-graph"
+    SERVER_VERSION = "1.0.0"
+    TOOLS = TOOLS  # 引用上方模块级 TOOLS (17个)
+
+    def handle_tool(self, name, args):
+        return _execute_tool(name, args)
 
 
 if __name__ == "__main__":
-    try:
-        main()
-    except KeyboardInterrupt:
-        pass
+    KnowledgeGraphMCPServer().run()