Skip to main content
Glama

Huasheng Shisan · Civil Service Exam Problem-Solving MCP Assistant

📚 Based on the Huasheng Shisan civil service exam knowledge base, 442 method cards covering four major modules: data analysis, quantitative relations, logical reasoning, and verbal comprehension. Paired with AI assistants (Claude Desktop, etc.), you can solve problems intelligently anytime, anywhere.

🚀 One-Minute Quick Start

Method 1: Let the Agent Install It for You (Easiest)

If you're a user of AI coding tools like Claude Code, Codex CLI, Cursor, Windsurf, just send the following sentence to it:

"Help me install this MCP server: https://github.com/heihei999/huasheng-mcp"

The Agent will automatically complete all the download, installation, and configuration steps. You don't have to do anything.

Method 2: Install It Yourself

Step 1: Install

pip install xingce-solver[sse]

If you don't have Python installed, go to https://www.python.org/downloads/ to download and install it, and check "Add Python to PATH".

Step 2: Start the Service

xingce-solver-mcp-sse

When you see Uvicorn running on http://0.0.0.0:8000, the service has started successfully 👌

Step 3: Connect to the AI Assistant

For Claude Desktop users, add the following to the configuration file (%APPDATA%\Claude\claude_desktop_config.json):

{
  "mcpServers": {
    "行测解题": {
      "type": "sse",
      "url": "http://localhost:8000/sse"
    }
  }
}

After saving, restart Claude Desktop, and you can directly call the civil service exam problem-solving tools in the chat!

ChatGPT / Other AI users: After starting the service, enter the address http://localhost:8000/sse in any MCP SSE-compatible client.


Related MCP server: Sequential-Thinking

📖 Knowledge Base Overview

The knowledge base contains 442 method cards in total, covering the four major modules of the civil service exam:

Module

Number of Cards

Description

📊 Data Analysis

52 cards

Growth rates, proportions, multiples, averages, etc.

🔢 Quantitative Relations

118 cards

Work problems, travel problems, permutations and combinations, etc.

🧠 Logical Reasoning

148 cards

Figure reasoning, logical judgment, definition judgment, analogy reasoning

💬 Verbal Comprehension

95 cards

Main idea and intent, sentence expression, logical fill-in-the-blank

🆕 Verbal Comprehension (New)

18 cards

Sentence ordering, detail judgment, etc.

🆕 Figure Reasoning (New)

10 cards

Figure pattern special topics

Total

442 cards

Covers all core question types


🛠️ Available Problem-Solving Tools (15 in total)

After connecting successfully, the AI assistant will automatically recognize the following tools:

Tool

What It Does

solve_data_analysis

📊 Solve data analysis questions

solve_logic_reasoning

🧠 Solve logical judgment questions

classify_question

🔍 Identify which module a question belongs to

search_methods

🔎 Search for problem-solving methods

get_method_card

📇 View the detailed content of a method

get_source_reference

📋 View the source of a method

route_xingce_question

🧭 Determine the question type and recommend a solution

compose_xingce_analysis_prompt

📝 Compose an analysis prompt

compose_xingce_answer_prompt

✅ Generate a conservative answer prompt

get_graphic_reasoning_scaffold

🎨 Figure reasoning method framework

get_definition_judgement_scaffold

📌 Definition judgment method framework

get_analogy_reasoning_scaffold

🔗 Analogy reasoning method framework

get_logic_analysis_scaffold

⚖️ Analytical reasoning method framework

get_quantity_relation_scaffold

🔢 Quantitative relations method framework

get_verbal_reasoning_scaffold

💬 Verbal comprehension method framework


🎯 Usage Examples in AI Assistants

All AI assistants that support MCP SSE can be connected (Claude Desktop, ChatGPT, Cherry Studio, etc.). Among them, Cherry Studio visualizes each tool call process, so beginners can intuitively see how the AI calls the knowledge base step by step to solve problems. It's especially suitable for those new to MCP.

Method 1: Send Text Questions Directly

After connecting, just type your question:

"Help me solve a data analysis question: In 2020, the revenue of a certain industry was 13.2 billion yuan, a year-over-year increase of 10%. What was the approximate revenue in 2019? A. 10 billion yuan B. 11 billion yuan C. 12 billion yuan D. 13.2 billion yuan"

The AI assistant will automatically call the tools and return the analysis process and answer.

Take a screenshot of a question from a practice app, PDF, or online course → send the screenshot directly to the AI assistant → the AI uses its "eyes" to read the image → automatically calls the problem-solving tools → instant answer

Any AI assistant that supports multimodal input can do this (Claude Desktop, ChatGPT, etc.). You don't need to type manually—just take a photo or screenshot. It's especially convenient for practicing on mobile or tablet.

Method 3: Search for Problem-Solving Methods

"Search for problem-solving methods about growth rate comparison"

The AI assistant will call the search_methods tool to find matching problem-solving techniques from the 442 method cards.

Method 4: Let the Agent Batch-Practice Questions

If you're using a coding agent like Claude Code or Codex, you can give it a path to process in batches:

"Read all the civil service exam question images in this folder, call the problem-solving tools one by one, and compile the answers into a table"

Great for intensive practice right before the exam.


📝 Battle-Tested Problem-Solving Prompt

A problem-solving SOP that the author has verified through real practice, which makes the AI strictly follow the "route → get scaffold → solve by rules" chain when calling MCP tools, preventing the AI from skipping the knowledge base and answering based on intuition. Copy the entire block below and send it to the AI assistant:

你是一个专业的公考行测名师。每次解答行测题目时,你必须严格遵循以下 SOP(标准操作流程) 链条来使用你的 MCP 知识库,绝不能跳过或卡死在任何一步:
【强制解题步骤】
第一步:提取与路由
当你收到用户的题目(或图片)时,你必须首先调用路由工具 route_xingce_question。
question_text 参数:仅填入你提取出来的纯净"行测题目原文"(如果有图片,必须提取图片中的完整题目文字,绝不能只填用户的闲聊!)。
module_hint 参数:如果用户指定了方法(如"十字交叉法"),必须填入此处;若无指定则留空。
第二步:获取脚手架
路由工具会返回一个结果,其中包含 recommended_tool(例如 get_quantity_relation_scaffold)。看清楚推荐的工具名后,你必须立刻去调用那个推荐的工具,从而获取详细的解题规则和公式!
注意:初次解答时禁止重复调用第一步的路由工具。但如果用户后续反驳、补充条件或提出新方法,允许你再次调用路由工具重新查阅资料!
第三步:按规则解答
拿到脚手架规则后,严格按照知识库给出的模板和步骤,结合用户指定的思路,为用户深度解答。
【异常兜底】
如果在第一步返回 "unknown" 或无匹配,或者在任何一步发生错误,只需尝试 1 次。如果依然失败,请立刻彻底放弃调用工具,直接动用你的内部推理能力为用户解答!

⚙️ Advanced Settings

Changing the Port

The default port is 8000. If it's occupied, you can change it:

set MCP_PORT=8080
xingce-solver-mcp-sse

LAN Sharing

To let other devices on the same LAN use it:

set MCP_HOST=0.0.0.0
xingce-solver-mcp-sse

Then other devices connect to http://your-IP-address:8000/sse.


🧪 Test Status

✅ 661 tests passed, 35 skipped, with full coverage of all 442 method cards in the knowledge base.


📋 Version History

Version

Highlights

v0.7.0

Knowledge base expanded from 292 to 442 cards, added SSE server, beginner-friendly

v0.6.0

Figure reasoning framework, 8 anti-patterns

v0.5.1

Module context boundary hardening, verified with 330 real exam questions

v0.5.0

Manual override for module context

v0.4.3

Conservative routing hardening, 57/60 on a 60-question stress test

v0.4.2

Data material signal recognition

v0.4.1

Answer gating security hardening

v0.4.0

First MCP integration version


💡 Questions? Submit them on GitHub Issues

Available Tools

15 tools
classify_questionC

Route a question stem to preliminary modules and method cards.

Inputs: question_text (required). Returns: object

compose_xingce_analysis_promptA

Compose a structured analysis prompt from question and route result. Supports module_hint / section_context to guide routing by exam section context. Does not solve questions, compute answers, or select options. Returns prompt_text for LLM consumption.

Inputs: question_text (required), options, module_hint, strict_mode, +3 more. Returns: object

compose_xingce_answer_promptB

Compose a conservative answer prompt for LLM-in-the-loop answering. Supports module_hint / section_context to guide routing by exam section context. Returns answer_prompt with strict constraints, output schema, and safety contract. Does not answer questions, call external LLM, or select options.

Inputs: question_text (required), options, module_hint, strict_mode, +7 more. Returns: object

get_analogy_reasoning_scaffoldA

Return the read-only method scaffold for analogy reasoning. This tool provides relation-type checklists, relation verification steps, option comparison guidance, and uncertainty policy. It does not solve questions or select an answer.

Inputs: no named parameters. Returns: object

get_definition_judgement_scaffoldA

Return the read-only method scaffold for definition judgement. This tool provides question polarity routing, definition element checklists, option verification template, and uncertainty policy. It does not solve questions or select an answer.

Inputs: no named parameters. Returns: object

get_graphic_reasoning_scaffoldA

Return the read-only method scaffold for graphic reasoning. This tool provides visual observation order, visual checklists, response template, and uncertainty policy. It does not inspect images, solve questions, or select an answer.

Inputs: no named parameters. Returns: object

get_logic_analysis_scaffoldA

Return the read-only method scaffold for logic analysis reasoning. This tool provides problem-type routing, structure templates, constraint extraction checklist, option verification guidance, and uncertainty policy. It does not solve questions or select an answer.

Inputs: no named parameters. Returns: object

get_method_cardC

Return a full method card by method_id.

Inputs: method_id (required). Returns: object

get_quantity_relation_scaffoldA

Return the read-only method scaffold for quantity relation reasoning. This tool provides problem-type routing, quantity extraction guidance, unit normalization checks, method checklists, option verification guidance, and uncertainty policy. It does not solve questions, compute final answers, or select an option.

Inputs: no named parameters. Returns: object

get_source_referenceB

Return source files, pages, confidence, and review flag for a method.

Inputs: method_id (required). Returns: object

get_verbal_reasoning_scaffoldA

Return the read-only method scaffold for verbal reasoning. This tool provides question-type routing, discourse-structure analysis guidance, cloze-context checks, sentence-expression checks, option verification guidance, and uncertainty policy. It does not solve questions, compute final answers, or select an option.

Inputs: no named parameters. Returns: object

route_xingce_questionA

Route a question to the recommended module or scaffold without solving. Supports module_hint / section_context to guide routing by exam section context. Returns module guess, confidence, recommended tool/track, and reasoning signals. Does not answer questions or select options.

Inputs: question_text (required), options, module_hint, strict_mode, +2 more. Returns: object

search_methodsC

Search method cards in the Xingce knowledge base.

Inputs: query (required), top_k, module. Returns: object

solve_data_analysisC

Build a structured data-analysis solving draft, not a final answer.

Inputs: question_text (required), options. Returns: object

solve_logic_reasoningC

Build a structured logic-reasoning solving draft, not a final answer.

Inputs: question_text (required), options. Returns: object

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 15 tools

    Disambiguation4/5

    Most tools have distinct purposes, but classify_question and route_xingce_question overlap slightly in routing, and compose_xingce_analysis_prompt vs compose_xingce_answer_prompt could confuse agents without careful reading. Overall, descriptions help disambiguate.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun snake_case pattern (e.g., classify_question, get_analogy_reasoning_scaffold, solve_data_analysis). No mixed conventions.

    Tool Count5/5

    With 15 tools, the set is well-scoped for the domain of question routing, scaffolding, and solving. Each tool serves a clear purpose, and the count is within the 3-15 ideal range.

    Completeness3/5

    The set covers routing, prompt composition, and scaffolds for six reasoning types, but only two solve draft tools (data analysis and logic reasoning) exist. Missing solve tools for analogy, definition, graphic, etc., leaving agents to work around these gaps.

    Maintenance

    ActivityStale
    ResponsivenessLimited data

    Related MCP Connectors

    Related MCP Servers