Skip to main content
Glama
K4G Application Triage MCP Server

K4G Application Triage MCP Server

by Sandypaluk

加速器申请筛选 MCP Server(演示版)

一个轻量级 MCP 服务器,可将创业公司/创始人申请的筛选流程转化为任何兼容 MCP 的 AI 客户端(如 Claude Desktop)都能用自然语言查询的内容,而无需手动翻查电子表格或表单工具。

该项目的初衷是面向食品系统和植物基创新领域的加速器——这类项目通常每一期都会运行一套结构化的申请、评分和导师评审流程。工具和模拟数据反映了这一背景,但底层模式是通用的:任何拥有申请流程的加速器、孵化器或资助项目,都可以将这个服务器指向自己的核心记录系统。

这是一个作品集 / 演示构建,因此这里的模拟数据所回答的问题专门针对该食品系统加速器背景(如 "alt-protein"、"fermentation" 等板块标签,以及食品行业的国家和创始人)。将 src/data.ts 换成另一份数据集,同一个四个工具就能回答金融科技加速器、气候科技项目或任何其他领域的同类问题——变化的是数据,而不是架构。在真实部署中,src/data.ts 将被替换为对组织实际使用的核心记录系统(Airtable、Notion、表单工具 API 等)的调用——下面这四个工具完全不需要改动。这正是 MCP 的意义所在:AI 客户端不关心工具背后是什么。

它可以回答的问题

  • "哪些 alt-protein 申请的 impact 评分达到 7 分以上?"

  • "David 的评审队列里还积压着什么?"

  • "还有哪些申请没有被任何评审人员接走?"

  • "调出 K4G-2026-058 的完整档案,包括备注。"

  • 将 Root & Policy Lab 标记为需要全体小组讨论——该团队需要进行联合创始人尽职调查。*

Related MCP server: TeamDesk MCP Server

工具

工具

功能

是否修改数据?

k4g_search_applications

按文本、行业、阶段或最低影响力评分筛选

否

k4g_get_application

单个申请的完整详情 + 评审备注

否

k4g_list_pending_review

列出所有仍需评审(可按评审人员筛选)的项目

否

k4g_flag_for_review

将申请标记为需进行小组讨论,并记录原因

是

设置

npm install
npm run build

独立运行

npm start

它会停留在 stdin/stdout 上等待 MCP 消息——这是正常的,它被设计为由 MCP 客户端启动,而非交互式运行。

连接到 Claude Desktop

  1. 打开 Claude Desktop 的配置文件:

    • macOS:~/Library/Application Support/Claude/claude_desktop_config.json

    • Windows:%APPDATA%\Claude\claude_desktop_config.json

  2. 添加此服务器(将路径替换为你放置此文件夹的位置):

    {
      "mcpServers": {
        "k4g-applications": {
          "command": "node",
          "args": ["/absolute/path/to/k4g-applications-mcp-server/dist/index.js"]
        }
      }
    }
  3. 重启 Claude Desktop。你应该会在工具/连接器图标下看到 "k4g-applications",并可以直接在聊天中提问。

向真实部署扩展

  • 将 src/data.ts 替换为真实的 API 客户端(Airtable/Notion 等)——src/index.ts 中的工具签名保持不变。

  • 谨慎添加写入工具:k4g_flag_for_review 刻意设计为唯一会修改数据的工具,并且它只做标记——绝不会拒绝或接受申请,最终决定权仍由人工掌控。

  • 如需托管/共享版本(支持多名评审人员,不仅限于本地使用),可将 stdio 传输替换为 Streamable HTTP——参见 https://modelcontextprotocol.io 上的 MCP 文档。

Available Tools

4 tools
k4g_flag_for_reviewFlag Application for Panel DiscussionA

Mark an application as 'flagged_for_discussion' and attach a reviewer note explaining why.

This changes the application's review_status - it is not read-only. Use it when an application needs a second opinion or a full-panel discussion rather than a solo decision.

Args:

  • id (string): application ID

  • reviewer (string): who is flagging it

  • note (string): reason for flagging

Returns the updated application record as JSON.

Error Handling:

  • Returns "Error: no application found with ID ''" if the ID doesn't exist.

Inputs: id (required), note (required), reviewer (required).

k4g_get_applicationGet K4G Application DetailA
Read-onlyIdempotent

Fetch full detail for a single application by ID, including all reviewer notes.

Args:

  • id (string): the application ID, e.g. "K4G-2026-014"

Returns the full application record as JSON, including the notes array.

Error Handling:

  • Returns "Error: no application found with ID ''" if the ID doesn't exist. Use k4g_search_applications to find the correct ID first.

Inputs: id (required).

k4g_list_pending_reviewList Applications Pending ReviewA
Read-onlyIdempotent

List applications that still need a decision: review_status is 'pending_review' or 'flagged_for_discussion'.

This is the tool to answer "what's stuck in the queue" - it does not include applications already marked 'reviewed'.

Args:

  • assigned_reviewer (string, optional): filter to one reviewer's queue

  • unassigned_only (boolean, optional): show only applications nobody owns yet (default false)

Returns JSON with: { count, applications: [ ...summaries, sorted oldest submitted_date first ] }

Examples:

  • "What's still in David's queue?" -> assigned_reviewer="David"

  • "What hasn't been picked up by anyone yet?" -> unassigned_only=true

Inputs: unassigned_only, assigned_reviewer.

k4g_search_applicationsSearch K4G ApplicationsA
Read-onlyIdempotent

Search the incubator application pipeline by text, sector, stage, or score.

This tool searches across all applications in the current cycle. It does NOT modify any data - use k4g_flag_for_review for that.

Args:

  • query (string, optional): matches against org name, founder name, one-liner

  • food_sector (string, optional): e.g. "alt-protein", "fermentation", "policy"

  • stage (enum, optional): pipeline stage to filter on

  • min_impact_score (number, optional): 1-10, minimum impact score

  • limit (number, optional): max results, default 20

Returns JSON with: { total, count, applications: [ { id, org_name, founder_name, country, food_sector, one_liner, stage, review_status, assigned_reviewer, impact_score, feasibility_score, team_score } ] }

Examples:

  • "Which alt-protein applications scored 7 or higher on impact?" -> food_sector="alt-protein", min_impact_score=7

  • "Find anything from Kenya" -> query="Kenya"

  • Don't use when: you already have the application ID (use k4g_get_application instead)

Inputs: limit, query, stage, food_sector, +1 more.

Tool Schema Changelog

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

Loading tool schema changelog…

    TDQS

    A4.4/5.0

    Scored across 4 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose: search finds applications by criteria, get fetches full details by ID, list surfaces pending decision items, and flag modifies review status. There is minor overlap between search and list, but the descriptions and intended use cases make the boundary explicit.

    Naming Consistency4/5

    All tools share the k4g_ prefix and use snake_case with a verb-first structure. k4g_flag_for_review and k4g_list_pending_review deviate slightly from the direct verb_noun pattern of the other two, but the convention is still clear and predictable.

    Tool Count5/5

    Four tools is a well-scoped set for an application triage server: search, detail view, pending queue, and flagging for discussion. Each tool covers a distinct step in the triage workflow without unnecessary bloat.

    Completeness3/5

    The read and flag workflows are covered, but there is no tool to mark an application as reviewed, assign a reviewer, or advance an application to the next pipeline stage. This creates a dead end where triage can only move applications into a discussion state, not out of it.

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      B
      maintenance
      An MCP server for Metrx — provides tools for construction, healthcare, logistics, manufacturing, and legal mid-market businesses to query and analyze their operational data via AI.
      42
      421 npm
      2
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      MCP server for TeamDesk databases, enabling CRUD operations, search, and document generation via natural language.
      4
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      MCP server for Stessa rental-property accounting platform, enabling AI agents to query properties, portfolios, banking, documents, and tenancies via natural language.
      MIT
    • F
      license
      A
      quality
      D
      maintenance
      MCP server that acts as a research intelligence agent, converting natural language queries into SQL or full-text search against Hacker News and arXiv data, and returning answers with supporting evidence.
      4
      -