mcp-gate: Instructions file for Claude Code

mcpbridge — AI Configuration Guide

This file tells you exactly how to configure mcpbridge for a user's site or application. Follow this precisely.

---

What mcpbridge does

mcpbridge is a zero-code MCP server runtime. One YAML file maps data sources — SQL queries, REST endpoints, and Markdown files — into MCP tools and resources that Claude can call. The operator writes YAML, not code. Restart to apply changes.

---

Your job when asked to configure mcpbridge

1. Ask the user: what data do they want Claude to access? (DB tables, APIs, docs?)
2. Choose the correct source type for each data source (see rules below)
3. Write a complete, valid config.yaml
4. Write a .env.example with every env var the config needs
5. Validate mentally: does each description tell Claude when to use this tool?

Never leave placeholders in the output. The config must be deployable as-is after the user fills in env vars.

---

Source type selection rules

What the user has	Use this type
A database table or query	db_query
A REST API endpoint (GET)	get_endpoint
Docs, FAQ, policies in .md files	knowledge_base
A website with no API	knowledge_base (scrape content to .md first, or ask user)

---

YAML schema — every field

server block

server:
  name: string           # kebab-case, e.g. "acme-mcp"
  version: "1.0.0"       # always start at 1.0.0
  transport: stdio       # always stdio unless user explicitly needs HTTP

db_query source

- type: db_query
  name: snake_case_name          # REQUIRED: snake_case only, no camelCase
  description: |                 # REQUIRED: min 20 chars, used by LLM to decide when to call
    One or two sentences.
    First sentence: what it returns.
    Second sentence: when to use it (trigger phrase).
  query: "SELECT ..."            # REQUIRED: parameterized with :param_name syntax
  params:                        # OPTIONAL: list of query parameters
    - name: param_name
      type: integer              # string | integer | boolean | float
      required: true             # or false
      default: value             # only if required: false
  connection: "${ENV_VAR_NAME}"  # REQUIRED: MUST use ${} syntax — never hardcode
  allow_write: false             # default false — only set true for explicit write ops