# mcp_quote_server
MCP server implementation

## Setup MCP configuration
Ensure the `.mcp.json` points to correct root folder of the project

## Part 0: Run MCP Server
```
uv run python quote_server/quote_mcp_server.py
```

## Part 1: The MCP server (no API key needed)

The server is [`quote_server/quote_mcp_server.py`](quote_server/quote_mcp_server.py):
~65 lines built on the MCP Python SDK's `FastMCP` helper. It reads
[`quotes.json`](quotes.json) and exposes two tools:

| Tool | Input | Returns |
| --- | --- | --- |
| `list_categories` | — | The categories that have quotes |
| `get_quote` | `category` | A random quote, or nothing if the category is unknown/empty |

### 1.1 — Smoke-test it (recommended)

[`try_server.py`](try_server.py) launches the server
over STDIO, speaks MCP to it, and prints a few sample calls. No LLM, no browser,
no Node — just the `mcp` SDK you already installed:

```bash
uv run python try_server.py
```

```
Tools: list_categories, get_quote
Categories: creativity, engineering, humor, motivation, stoicism
A stoic quote: “We suffer more often in imagination than in reality.” — Seneca
Unknown category 'banana': '' (nothing, as expected)
```

That's a complete, working MCP server. Everything below is just pointing
different hosts at it.

### 1.2 — Optional: the MCP Inspector

The [MCP Inspector](https://github.com/modelcontextprotocol/inspector) is
Anthropic's official tool for poking at servers (needs Node/npx). It has two
modes.

**CLI mode (recommended — no browser, scriptable):**

```bash
# list the tools
npx @modelcontextprotocol/inspector --cli \
  uv run python quote_server/quote_mcp_server.py \
  --method tools/list

# call a tool
npx @modelcontextprotocol/inspector --cli \
  uv run python quote_server/quote_mcp_server.py \
  --method tools/call --tool-name get_quote --tool-arg category=stoicism
```

The second command prints the quote as JSON. (If your npm registry rejects
scoped packages, add `--registry=https://registry.npmjs.org` to `npx`.)

**GUI mode:**

Point the Inspector at our `.mcp.json` (the same file Claude Code uses) and
name the server — it opens with the STDIO transport and command pre-filled:

```bash
npx @modelcontextprotocol/inspector --config .mcp.json --server quotes
```

A browser tab opens with **Transport Type: STDIO**, **Command: `uv`**, and our
arguments already populated. Click **Connect** (top-left) — the status flips to
**Connected** — then open the **Tools** tab, **List Tools**, and run them.

(The committed `.mcp.json` has an absolute `--directory` path. If you cloned the
repo somewhere other than `~/git/summit-ai-mcp-demo`, edit that path first — or
regenerate the file with the Claude Code step in [3.1](#31--claude-code).)

## Part 2: Claude Code

```bash
claude mcp add quotes --scope project \
  -- uv --directory ~/Dev/Python/mcp_quote_server run python quote_server/quote_mcp_server.py
```

This writes a `.mcp.json` in the repo (already committed here as a reference).
Run `claude` in the repo and approve the project server when prompted, then:

```
> give me a quote about creativity
```

Verify anytime with inside claude with  `/mcp`

## Part 3: Graphify the codebase

Use lower-cost model like minimax3 using Ollama in Claude code
```bash
ollama launch claude --model minimax-m3:cloud
```

Graphify the codebase inside ~/Dev/Python/mcp_quote_server, using skill
```
/graphify .
```

The code-base graph would be available inside folder `graphify-out`