
GPT-5 MCP Server
Enables GPT-5 inference via OpenAI API with configurable reasoning effort, verbosity levels, and optional web search integration. Supports per-call parameter overrides for flexible AI model interactions through MCP.
README
GPT-5 MCP Server (TypeScript)
An MCP server that exposes a gpt5_query
tool for GPT-5 inference via OpenAI Responses API, with optional Web Search Preview. Supports per-call overrides for verbosity, reasoning effort, and other parameters.
Features
- TypeScript MCP server using
@modelcontextprotocol/sdk
gpt5_query
toolweb_search_preview
integration (optional)verbosity
(low|medium|high)reasoning.effort
(low|medium|high)tool_choice
(auto|none),parallel_tool_calls
system
prompt,model
,max_output_tokens
- Config via environment variables with per-call overrides
Quick Start
- Install dependencies
pnpm i # or npm i / yarn
- Configure environment
Create .env
(or export env vars):
OPENAI_API_KEY=sk-...
OPENAI_MODEL=gpt-5
OPENAI_MAX_RETRIES=3
OPENAI_TIMEOUT_MS=60000
REASONING_EFFORT=medium
DEFAULT_VERBOSITY=medium
WEB_SEARCH_DEFAULT_ENABLED=false
WEB_SEARCH_CONTEXT_SIZE=medium
- Build and run
pnpm run build
pnpm start
For development (watch mode):
pnpm run dev
Using with MCP Clients (Claude Code, Claude Desktop)
This server speaks Model Context Protocol (MCP) over stdio and emits pure JSON to stdout, making it safe for Claude Code and Claude Desktop.
Prerequisites
- Node.js 18+
- OpenAI API key via
.env
or environment variable
- Build
pnpm run build
- Run directly (recommended)
- Command:
node
- Args:
dist/cli.js
- CWD: repository root (required if you want
.env
to be loaded)
Example:
node dist/cli.js
- Add to Claude Code (VS Code)
- Command Palette → "Claude: Manage MCP Servers"
- "Add server" with:
- Name:
gpt5-mcp
- Command:
node
(or absolute path, e.g.,/opt/homebrew/bin/node
) - Args: ["/absolute/path/to/gpt5-mcp-server/dist/cli.js"] (or just
gpt5-mcp-server
if installed globally) - Env (choose one):
- Option A (ENV_FILE):
ENV_FILE=/absolute/path/to/gpt5-mcp-server/.env
- Option B (explicit): set
OPENAI_API_KEY
,OPENAI_MODEL
,OPENAI_TIMEOUT_MS
,DEFAULT_VERBOSITY
,REASONING_EFFORT
,WEB_SEARCH_DEFAULT_ENABLED
,WEB_SEARCH_CONTEXT_SIZE
- Option A (ENV_FILE):
- Name:
- Add to Claude Desktop
Edit config (e.g., macOS:
~/Library/Application Support/Claude/claude_desktop_config.json
) and add:
Option A: using ENV_FILE
{
"mcpServers": {
"gpt5-mcp": {
"command": "/opt/homebrew/bin/node",
"args": ["/absolute/path/to/gpt5-mcp-server/dist/cli.js"],
"env": {
"ENV_FILE": "/absolute/path/to/gpt5-mcp-server/.env"
}
}
}
}
Option B: explicit env vars
{
"mcpServers": {
"gpt5-mcp": {
"command": "/opt/homebrew/bin/node",
"args": ["/absolute/path/to/gpt5-mcp-server/dist/cli.js"],
"env": {
"OPENAI_API_KEY": "sk-...",
"OPENAI_MODEL": "gpt-5",
"OPENAI_TIMEOUT_MS": "120000",
"DEFAULT_VERBOSITY": "medium",
"REASONING_EFFORT": "low",
"WEB_SEARCH_DEFAULT_ENABLED": "false",
"WEB_SEARCH_CONTEXT_SIZE": "medium"
}
}
}
}
- CLI usage
- Package exposes bin(s).
- Local link:
npm link
→ rungpt5-mcp-server
- Global (after publish):
npm i -g gpt5-mcp-server
→gpt5-mcp-server
- Direct:
node /absolute/path/to/gpt5-mcp-server/dist/cli.js
- Local link:
- Web Search notes
- Due to OpenAI constraints,
web_search_preview
cannot be combined withreasoning.effort = minimal
. - This server automatically bumps effort to
medium
ifweb_search.enabled = true
. - If you need strict
minimal
, setweb_search.enabled = false
.
- Troubleshooting
- JSON parse error (Unexpected token ...)
- Likely extra logs on stdio. Use
node dist/cli.js
, avoidnpx
.
- Likely extra logs on stdio. Use
- Auth error
- Ensure
OPENAI_API_KEY
is provided.
- Ensure
- Timeout
- Increase
OPENAI_TIMEOUT_MS
(e.g., 120000).
- Increase
- 400 with Web Search
- Caused by
minimal
effort + web search. It's auto-bumped tomedium
; alternatively setreasoning_effort=medium
or disableweb_search
.
- Caused by
Tool: gpt5_query
Input schema (JSON):
{
"query": "string",
"model": "string?",
"system": "string?",
"reasoning_effort": "low|minimal|medium|high?",
"verbosity": "low|medium|high?",
"tool_choice": "auto|none?",
"parallel_tool_calls": "boolean?",
"max_output_tokens": "number?",
"web_search": {
"enabled": "boolean?",
"search_context_size": "low|medium|high?"
}
}
Example call (Inspector or client):
{
"method": "tools/call",
"params": {
"name": "gpt5_query",
"arguments": {
"query": "Summarize the latest on X.",
"verbosity": "low",
"web_search": { "enabled": true, "search_context_size": "medium" }
}
}
}
Defaults and behavior
- model: defaults to
OPENAI_MODEL
(env). Example:gpt-5
. - system: optional. Sent as
instructions
. - reasoning_effort: accepts
low|minimal|medium|high
. Internallylow
→minimal
。- Constraint: when
web_search.enabled=true
and effort isminimal
, it is auto-bumped tomedium
to satisfy OpenAI constraints.
- Constraint: when
- verbosity: defaults to
DEFAULT_VERBOSITY
(env). Sent astext.verbosity
. - tool_choice: default
auto
. - parallel_tool_calls: default
true
. - max_output_tokens: optional; omitted when not set.
- web_search.enabled: defaults to
WEB_SEARCH_DEFAULT_ENABLED
(env). - web_search.search_context_size: defaults to
WEB_SEARCH_CONTEXT_SIZE
(env). Allowed:low|medium|high
.
Environment variable mapping
OPENAI_API_KEY
(required)OPENAI_MODEL
→ model defaultOPENAI_MAX_RETRIES
→ OpenAI clientOPENAI_TIMEOUT_MS
→ OpenAI clientREASONING_EFFORT
→ reasoning_effort default (low|minimal|medium|high
)DEFAULT_VERBOSITY
→ verbosity default (low|medium|high
)WEB_SEARCH_DEFAULT_ENABLED
→ web_search.enabled default (true|false
)WEB_SEARCH_CONTEXT_SIZE
→ web_search.search_context_size default (low|medium|high
)
Output shape
- On success:
content: [{ type: "text", text: string }]
- On error:
isError: true
and atext
item withError: ...
Notes
- If the selected model does not support certain fields (e.g.,
verbosity
), they are ignored. - Keep API keys out of logs. Ensure
.env
is not committed.
License
MIT
日本語 (Japanese)
<a id="ja"></a>
ツール: gpt5_query
入力スキーマ (JSON):
{
"query": "string",
"model": "string?",
"system": "string?",
"reasoning_effort": "low|minimal|medium|high?",
"verbosity": "low|medium|high?",
"tool_choice": "auto|none?",
"parallel_tool_calls": "boolean?",
"max_output_tokens": "number?",
"web_search": {
"enabled": "boolean?",
"search_context_size": "low|medium|high?"
}
}
例 (Inspector など):
{
"method": "tools/call",
"params": {
"name": "gpt5_query",
"arguments": {
"query": "Summarize the latest on X.",
"verbosity": "low",
"web_search": { "enabled": true, "search_context_size": "medium" }
}
}
}
既定値と挙動
- model: 既定は
OPENAI_MODEL
(環境変数)。例:gpt-5
。 - system: 任意。OpenAI には
instructions
として送信します。 - reasoning_effort:
low|minimal|medium|high
を受け付け、内部的にlow
はminimal
として扱われます。- 制約:
web_search.enabled=true
かつ effort=minimal
の場合、OpenAI の制約に合わせて自動的にmedium
に引き上げます。
- 制約:
- verbosity: 既定は
DEFAULT_VERBOSITY
(環境変数)。OpenAI にはtext.verbosity
として送信します。 - tool_choice: 既定は
auto
。 - parallel_tool_calls: 既定は
true
。 - max_output_tokens: 任意。未指定の場合は送信しません。
- web_search.enabled: 既定は
WEB_SEARCH_DEFAULT_ENABLED
(環境変数)。 - web_search.search_context_size: 既定は
WEB_SEARCH_CONTEXT_SIZE
(環境変数)。許容値:low|medium|high
。
環境変数マッピング
OPENAI_API_KEY
(必須)OPENAI_MODEL
→ model 既定OPENAI_MAX_RETRIES
→ OpenAI クライアント設定OPENAI_TIMEOUT_MS
→ OpenAI クライアント設定REASONING_EFFORT
→ reasoning_effort 既定(low|minimal|medium|high
)DEFAULT_VERBOSITY
→ verbosity 既定(low|medium|high
)WEB_SEARCH_DEFAULT_ENABLED
→ web_search.enabled 既定(true|false
)WEB_SEARCH_CONTEXT_SIZE
→ web_search.search_context_size 既定(low|medium|high
)
出力形式
- 成功時:
content: [{ type: "text", text: string }]
- エラー時:
isError: true
とtext
にError: ...
注意
- 選択したモデルが特定のフィールド(例:
verbosity
)をサポートしない場合、それらは無視されます。 - API キーはログに出力しません。
.env
はコミットしないでください。
MCP Serverの使い方
このサーバーは Model Context Protocol (MCP) の標準入出力(stdio)で動作します。純粋な JSON のみを stdout に出力する設計のため、MCP Inspector / Claude Code / Claude Desktop で安全に接続できます。
前提
- Node.js 18+
- OpenAI APIキーが
.env
もしくは環境変数で設定されていること
- ビルド
pnpm run build
- 直接起動(推奨)
- コマンド:
node
- 引数:
dist/cli.js
- CWD: リポジトリのルート(
.env
を読む場合は必須)
例:
node dist/cli.js
- Claude Code(VS Code 拡張)に追加
- VS Code のコマンドパレット → 「Claude: Manage MCP Servers」
- 「Add server」で次を入力:
- Name:
gpt5-mcp
- Command:
node
(絶対パス可) - Args:
["/絶対/パス/gpt5-mcp-server/dist/cli.js"]
(グローバル導入済みなら不要) - Env(どちらか一方):
- オプションA(ENV_FILE):
ENV_FILE=/絶対/パス/gpt5-mcp-server/.env
- オプションB(明示指定):
OPENAI_API_KEY
、OPENAI_MODEL
、OPENAI_TIMEOUT_MS
、DEFAULT_VERBOSITY
、REASONING_EFFORT
、WEB_SEARCH_DEFAULT_ENABLED
、WEB_SEARCH_CONTEXT_SIZE
- オプションA(ENV_FILE):
- Name:
- Claude Desktop に追加
設定ファイル(例: macOS は
~/Library/Application Support/Claude/claude_desktop_config.json
)を編集して以下を追記します。
オプションA: ENV_FILE を使う
{
"mcpServers": {
"gpt5-mcp": {
"command": "/opt/homebrew/bin/node",
"args": ["/絶対/パス/gpt5-mcp-server/dist/cli.js"],
"env": {
"ENV_FILE": "/絶対/パス/gpt5-mcp-server/.env"
}
}
}
}
オプションB: 環境変数を明示指定
{
"mcpServers": {
"gpt5-mcp": {
"command": "/opt/homebrew/bin/node",
"args": ["/絶対/パス/gpt5-mcp-server/dist/cli.js"],
"env": {
"OPENAI_API_KEY": "sk-...",
"OPENAI_MODEL": "gpt-5",
"OPENAI_TIMEOUT_MS": "120000",
"DEFAULT_VERBOSITY": "medium",
"REASONING_EFFORT": "low",
"WEB_SEARCH_DEFAULT_ENABLED": "false",
"WEB_SEARCH_CONTEXT_SIZE": "medium"
}
}
}
}
- CLI の利用
- パッケージには bin が含まれます。
- ローカルリンク:
npm link
後にgpt5-mcp-server
- グローバル(公開後):
npm i -g gpt5-mcp-server
→gpt5-mcp-server
- 直接実行:
node /絶対/パス/gpt5-mcp-server/dist/cli.js
- ローカルリンク:
- Web Search に関する注意
- OpenAI の制約により、
web_search_preview
はreasoning.effort = minimal
と併用できません。 - 本サーバーは
web_search.enabled = true
の場合、自動的に effort をmedium
に引き上げて呼び出します。 - もし
minimal
を厳格に使いたい場合は、web_search.enabled = false
にしてください。
- トラブルシューティング
- JSON パースエラー(Unexpected token ...)
- stdio に余計な出力が混ざっている可能性があります。
node dist/cli.js
を使い、npx
は避けてください。 .env
読み込みやライブラリのログは既に抑止済みです。
- stdio に余計な出力が混ざっている可能性があります。
- 認証エラー
OPENAI_API_KEY
が正しく渡っているか確認。
- タイムアウト
OPENAI_TIMEOUT_MS
を増やす(例: 120000)。
- Web Search で 400 エラー
reasoning.effort=minimal
とweb_search
の併用不可が原因。自動的にmedium
に上げますが、明示的にmedium
を指定するか、web_search
を無効化してください。
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.