uni-kb
Enables parsing, indexing, and querying source code as structured knowledge, providing code exploration, spec generation, and migration tools via 20 MCP tools.
README
uni-kb — Knowledge Base Library
Reusable data layer for codebase understanding. Parses source code into structured knowledge, indexes it for search, and serves it via MCP.
Can be used standalone by human developers or consumed by uni-dev agent system.
Architecture
Source Code
│
▼
┌──────────────────────────────────────────┐
│ tree-sitter Parsers (plugin system) │
│ Java/Spring · Node.js/Express/NestJS │
└───────────────┬──────────────────────────┘
▼ structured JSON
┌──────────────────────────────────────────┐
│ Storage Layer │
│ ┌──────────┬──────────┬───────────────┐ │
│ │ SQLite │ ChromaDB │ NetworkX │ │
│ │ 8 tables │ 14 idx │ Code Graph │ │
│ └──────────┴──────────┴───────────────┘ │
└───────────────┬──────────────────────────┘
▼
┌──────────────────────────────────────────┐
│ Spec Generators (6) │
│ API Contract · Business Logic · Models │
│ Auth Matrix · Config · Migration │
└───────────────┬──────────────────────────┘
▼
┌──────────────────────────────────────────┐
│ MCP Server (20 tools, 5 categories) │
└──────────────────────────────────────────┘
Phase 1: Parser Extension System
Goal: Abstract plugin interface + registry + Java parser
1.1 Base Plugin Interface (parsers/base.py)
class ParserPlugin(ABC):
@abstractmethod
def language(self) -> str: ...
@abstractmethod
def detect(self, file_path: str) -> bool: ...
@abstractmethod
def parse(self, file_path: str, source: str) -> ParseResult: ...
1.2 Plugin Registry (parsers/registry.py)
Discovers plugins via Python entry points (pyproject.toml [project.entry-points."uni_kb.parsers"]). Loads on demand.
1.3 Java Parser (parsers/java/)
Four sub-parsers using tree-sitter-java:
| Parser | Target | Extracts |
|---|---|---|
controller.py |
@RestController, @RequestMapping |
Endpoints, HTTP methods, paths, params, response types |
service.py |
@Service, method bodies |
Business logic structure, dependencies |
entity.py |
@Entity, @Column, @Table |
Entities, fields, types, constraints, relationships |
mapper.py |
MyBatis XML | Custom SQL queries, result maps |
Output: ParseResult with standardized JSON matching SQLite schema.
Phase 2: Storage Layer
2.1 SQLite Store (store/sqlite_store.py)
8 tables initialized on uni-kb init --project ./:
| Table | Purpose |
|---|---|
modules |
Top-level source modules |
classes |
Classes / components |
methods |
Methods / functions |
api_endpoints |
REST endpoints |
db_tables |
Database tables |
db_columns |
Table columns |
frontend_components |
Vue/Cocos components |
auth_permissions |
Permission annotations |
2.2 ChromaDB Indexes (store/chroma_indexes.py)
14 named collections with embedded mode (persisted in .uni-kb/chroma/):
| # | Index | Content |
|---|---|---|
| 1 | code_java_controller |
Java controller source |
| 2 | code_java_service |
Java service source |
| 3 | code_java_entity |
Java entity source |
| 4 | code_java_mapper |
Mapper XML source |
| 5 | code_typescript_admin |
Vue admin source |
| 6 | code_typescript_game |
Cocos game source |
| 7 | specs_api |
Generated OpenAPI specs |
| 8 | specs_business_logic |
Generated method docs |
| 9 | specs_data_model |
Generated entity specs |
| 10 | specs_auth |
Permission matrix |
| 11 | specs_config |
Config catalog |
| 12 | db_schema |
Table DDL |
| 13 | project_docs |
Documentation |
| 14 | migration_checklists |
Per-module checklists |
2.3 Code Graph (store/code_graph.py)
NetworkX directed graph.
Nodes (8 types): Module, Class, Method, APIEndpoint, DBTable, DBColumn, VueComponent, CocosComponent
Edges (11 + 3 migration): HAS_METHOD, CALLS, INJECTS, IMPLEMENTS, EXTENDS, ROUTES_TO, MAPS_TO, FK_TO, API_CALLER, PERMITS, IMPORTS + MIGRATES_TO, BLOCKED_BY, VERIFIED_AGAINST
Phase 3: Node.js Parser
1.4 Node.js Parser (parsers/nodejs/)
Four sub-parsers using tree-sitter-typescript:
| Parser | Target | Extracts |
|---|---|---|
route.py |
Express router.get() / NestJS @Get() |
Endpoints, HTTP methods, paths, middleware chains |
service.py |
Service classes | Business logic structure |
model.py |
Sequelize / TypeORM models | Entities, fields, types, relationships |
middleware.py |
Auth middleware | Permission chains, guards |
Phase 4: Spec Generators + MCP
4.1 Generators (generators/)
6 generators, 4 exposed as direct MCP tools:
| Generator | Input | Output | MCP Tool |
|---|---|---|---|
api_contract.py |
Controller/Route AST | OpenAPI 3.0 YAML | get_api_contract |
business_logic.py |
Service AST | Markdown pseudo-code | get_business_logic_doc |
data_model.py |
Entity + DB schema | YAML model spec | → get_entity_spec |
auth_matrix.py |
Permission annotations | Permission matrix YAML | get_permission_matrix |
config_catalog.py |
YAML/.env files | Config catalog YAML | → get_config_value |
migration_checklist.py |
Module + dep graph | Prioritized checklist MD | get_migration_checklist |
4.2 MCP Server (mcp_server.py)
20 tools in 5 categories:
| Category | Tools |
|---|---|
| Code Exploration | search_code, get_method_body, get_class_structure, find_endpoints, find_usages |
| Spec Retrieval | get_api_contract, get_business_logic_doc, verify_contract, compare_api_responses |
| Data Model | get_entity_spec, get_db_schema, get_column_info, trace_fk_chain |
| Auth & Config | get_permission_matrix, get_config_value, get_config_catalog |
| Operations | get_migration_checklist, get_dependency_graph, get_migration_status, run_test_suite |
Dependencies
chromadb
networkx
tree-sitter
tree-sitter-java
tree-sitter-typescript
sqlite-utils
pyyaml
mcp>=1.0.0
CLI Usage
# Initialize knowledge base for a project
uni-kb init --project /path/to/project
# Parse and index source code
uni-kb index --project /path/to/project
# Start MCP server
uni-kb serve --project /path/to/project --port 9020
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。