Skip to main content
Glama
AI App MCP

AI App MCP

by yh008

AI App MCP

A production-ready Python MCP server that supports both stdio and Streamable HTTP transports.

Features

  • health_check: return server health and runtime configuration.

  • normalize_user_query: normalize user text before retrieval or agent routing.

  • search_knowledge_base: search local .md, .txt, and .json files.

  • get_document: read a safe document from the configured knowledge directory.

  • build_rag_prompt: build a grounded RAG prompt.

  • config://runtime: expose safe runtime configuration.

  • rag_answer_prompt: reusable RAG prompt template.

Related MCP server: Python MCP Server Template

Install

cd D:\projects\codex\single\mcp
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -e ".[dev]"
Copy-Item .env.example .env

Transport 1: stdio

Use stdio when an AI client starts this MCP server as a local child process.

ai-app-mcp --transport stdio

Equivalent module command:

python -m ai_app_mcp.server --transport stdio

Client configuration example:

{
  "mcpServers": {
    "ai-app-mcp": {
      "command": "python",
      "args": [
        "-m",
        "ai_app_mcp.server",
        "--transport",
        "stdio"
      ],
      "env": {
        "MCP_KNOWLEDGE_DIR": "D:/projects/codex/single/mcp/knowledge"
      }
    }
  }
}

In stdio mode, stdout is used for MCP protocol messages. Logs are written to stderr.

Transport 2: Streamable HTTP

Use Streamable HTTP for MCP Inspector, HTTP debugging, or service-to-service integration.

ai-app-mcp --transport streamable-http --host 127.0.0.1 --port 8000

Equivalent module command:

python -m ai_app_mcp.server --transport streamable-http --host 127.0.0.1 --port 8000

MCP endpoint:

http://127.0.0.1:8000/mcp

Do not open this endpoint directly as a normal web page. It is a JSON-RPC MCP endpoint and requires an MCP client.

Debug With MCP Inspector

Start this MCP server:

ai-app-mcp --transport streamable-http --host 127.0.0.1 --port 8000

Start Inspector:

npx -y @modelcontextprotocol/inspector

Open:

http://localhost:6274

Use:

Transport: Streamable HTTP
URL: http://127.0.0.1:8000/mcp

Environment

Copy .env.example to .env and adjust values:

MCP_SERVER_NAME=ai-app-mcp
MCP_LOG_LEVEL=INFO
MCP_KNOWLEDGE_DIR=./knowledge
MCP_MAX_TEXT_CHARS=12000

Knowledge Directory

By default, the server reads files from ./knowledge. Only .md, .txt, and .json files are allowed. Paths are resolved safely so clients cannot read files outside the knowledge directory.

Test

pytest

Available Tools

5 tools
build_rag_promptB

Build a grounded RAG prompt from a question and retrieved context.

Inputs: context (required), question (required), answer_style. Returns: result

get_documentB

Read a document from the configured knowledge directory.

Inputs: relative_path (required). Returns: object

health_checkA

Return server health and runtime configuration.

Inputs: no named parameters. Returns: object

normalize_user_queryC

Normalize user text before retrieval, routing, or agent planning.

Inputs: query (required), lowercase. Returns: object

search_knowledge_baseC

Search local knowledge files by keyword.

Inputs: query (required), limit. Returns: result

Tool Schema Changelog

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

Loading tool schema changelog…

    TDQS

    B3.4/5.0

    Scored across 5 tools

    Disambiguation4/5

    All tools have distinct roles: search and get_document are clearly separate (search vs. reading full doc), build_rag_prompt and normalize_user_query serve different pipeline stages, and health_check is standalone. Slight potential confusion exists between search_knowledge_base and get_document, but descriptions resolve this.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun snake_case pattern: search_knowledge_base, get_document, build_rag_prompt, normalize_user_query, health_check. The pattern is predictable and readable.

    Tool Count5/5

    Five tools is a well-scoped count for a RAG-focused server. Each tool serves a clear purpose in the pipeline, and the count feels neither too thin nor overly heavy.

    Completeness4/5

    The set covers the core RAG flow (normalize, search, retrieve, build prompt) plus health check. Missing operations like adding/updating knowledge base documents or a combined retrieve tool, but these are minor gaps for the apparent scope.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      A production-grade MCP server and client implementation with comprehensive features including structured logging, health checks, metrics, authentication, and RAG capabilities with PostgreSQL vector search. Supports both stdio and SSE transports with containerization and security features for enterprise deployment.
      MIT
    • F
      license
      Not graded
      quality
      D
      maintenance
      A foundational template for building MCP servers in Python using Streamable HTTP transport. Provides example implementations of tools, resources, and prompts to help developers create custom MCP integrations for AI assistants.
      -
    • A
      license
      Not graded
      quality
      D
      maintenance
      A comprehensive Python MCP server with built-in knowledge base (SQLite + FTS5), web management interface, and flexible tool grouping system. Supports multiple transport protocols (stdio, SSE, HTTP Stream) with zero external dependencies.
      MIT