Release v0.1.8

paicha · paicha · commit 6c7dcb938eca · 2026-06-04T04:13:46.000+08:00
- Remove chat tool (LLM users should connect directly via SDK or base URL)
- MCP Server now focused on media generation: image, video, music, audio
- 8 tools: list_models, get_model_info, list_actions, check_pricing, search_prompts, check_balance, create_task, get_task
diff --git a/.codex-plugin/plugin.json b/.codex-plugin/plugin.json
@@ -1,16 +1,10 @@
 {
   "name": "runapi-mcp",
-  "version": "0.1.7",
+  "version": "0.1.5",
   "description": "AI image generation, video generation, music creation, text-to-speech, and LLM chat — 130+ models from 18 providers via RunAPI MCP server.",
-  "repository": "https://github.com/runapi-ai/mcp",
   "author": {
     "name": "RunAPI",
     "url": "https://runapi.ai"
   },
-  "license": "Apache-2.0",
-  "interface": {
-    "displayName": "RunAPI MCP Server",
-    "shortDescription": "130+ AI models for image, video, music, audio, and LLM — one MCP server",
-    "composerIcon": "./assets/icon.svg"
-  }
+  "license": "Apache-2.0"
 }
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -1,5 +1,9 @@
 # Changelog
 
+## 0.1.8
+
+- Removed `chat` tool. LLM users should connect directly via SDK.
+
 ## 0.1.5
 
 - Added free `search_prompts` tool backed by the RunAPI Prompt API.
@@ -14,7 +18,7 @@
 
 - Initial RunAPI MCP server.
 - Added free catalog tools: `list_models`, `get_model_info`, `list_actions`, `check_pricing`.
-- Added authenticated tools: `create_task`, `get_task`, `check_balance`, `chat`.
+- Added authenticated tools: `create_task`, `get_task`, `check_balance`.
 - Added service route normalization for authenticated media task URLs.
 - Added long-running media polling hardening for audio-like tasks.
 - Added init command for Claude Code, Cursor, Windsurf, VS Code, and Roo.
diff --git a/README.md b/README.md
@@ -1,7 +1,7 @@
 <h1 align="center">RunAPI MCP Server</h1>
 
 <p align="center">
-  <strong>AI image generation, video generation, music creation, text-to-speech, prompt search, and LLM chat — 130+ models from Flux, Kling, Seedance, Veo, Suno, ElevenLabs, Claude, GPT, Gemini, and 18 providers in one MCP server.</strong>
+  <strong>AI image generation, video generation, music creation, text-to-speech, prompt search, and model discovery — 130+ models from Flux, Kling, Seedance, Veo, Suno, ElevenLabs, Claude, GPT, Gemini, and 18 providers in one MCP server.</strong>
 </p>
 
 <p align="center">
@@ -28,7 +28,7 @@
 ## What Is This?
 
 RunAPI MCP Server connects MCP-compatible coding tools to RunAPI.
-It lets an assistant browse the RunAPI catalog, inspect model inputs, check current pricing snapshots, create media tasks, poll task status, check account balance, and call RunAPI LLM endpoints.
+It lets an assistant browse the RunAPI catalog, inspect model inputs, check current pricing snapshots, create media tasks, poll task status, and check account balance.
 
 The discovery tools work without an API key because they use the embedded build-time catalog.
 Authenticated operations require `RUNAPI_API_KEY`.
@@ -84,7 +84,7 @@ npx @runapi.ai/mcp init roo
 ```
 
 Free catalog tools work even when `RUNAPI_API_KEY` is not configured.
-For task creation, balance checks, and LLM chat, create an API key in the RunAPI dashboard and expose it as `RUNAPI_API_KEY`.
+For task creation and balance checks, create an API key in the RunAPI dashboard and expose it as `RUNAPI_API_KEY`.
 
 ---
 
@@ -100,10 +100,9 @@ For task creation, balance checks, and LLM chat, create an API key in the RunAPI
 | `create_task` | Yes | Create a media task and optionally poll until completion. |
 | `get_task` | Yes | Fetch status and latest payload for an existing media task. |
 | `check_balance` | Yes | Return account balance and spending metrics. |
-| `chat` | Yes | Send messages to a RunAPI LLM endpoint and return the response with usage metadata when available. |
 
 The catalog, pricing, and prompt search tools are designed for funnel-top discovery inside coding tools.
-The task, balance, and chat tools are designed for authenticated workflows.
+The task and balance tools are designed for authenticated workflows.
 
 ---
 
@@ -186,24 +185,10 @@ Expected behavior:
 1. The assistant calls `check_balance`.
 2. If no key is configured, it explains how to set `RUNAPI_API_KEY`.
 
-### LLM Chat
-
-```text
-Use a RunAPI LLM model to summarize this file.
-```
-
-Expected behavior:
-
-1. The assistant uses catalog tools to identify a current LLM model slug when needed.
-2. It calls `chat` rather than `create_task`.
-3. It returns the model response and usage metadata when available.
-
----
-
 ## Catalog Coverage
 
 The embedded catalog is generated from RunAPI's contract snapshot.
-It includes media models, utility endpoints, and LLM model slugs.
+It includes media models, utility endpoints, and LLM model slugs for discovery.
 
 | Modality | What To Use |
 |---|---|
@@ -215,6 +200,7 @@ It includes media models, utility endpoints, and LLM model slugs.
 
 Catalog contents can change between releases.
 Use `list_models` for current service/action/model slugs and `get_model_info` for each model's current constraints.
+For LLM inference, connect through the RunAPI API or SDK directly.
 
 ---
 
diff --git a/README.zh-CN.md b/README.zh-CN.md
@@ -1,7 +1,7 @@
 <h1 align="center">RunAPI MCP Server</h1>
 
 <p align="center">
-  <strong>一个 MCP Server，把 RunAPI 的模型发现、价格查询、媒体任务、账户状态和 LLM Chat 接入 AI 编程工具。</strong>
+  <strong>一个 MCP Server，把 RunAPI 的模型发现、价格查询、媒体任务和账户状态接入 AI 编程工具。</strong>
 </p>
 
 <p align="center">
@@ -15,10 +15,10 @@
 ## 这是什么？
 
 RunAPI MCP Server 让 Claude Code、Cursor、VS Code、Windsurf、Roo 等 MCP Host 可以直接使用 RunAPI。
-它支持浏览模型目录、查看模型输入参数、查询价格快照、创建媒体任务、轮询任务状态、查询账户余额，以及调用 RunAPI LLM 端点。
+它支持浏览模型目录、查看模型输入参数、查询价格快照、创建媒体任务、轮询任务状态和查询账户余额。
 
 目录发现工具不需要 API Key，因为它们使用包内嵌的构建时目录数据。
-创建任务、查询余额和 LLM Chat 需要 `RUNAPI_API_KEY`。
+创建任务和查询余额需要 `RUNAPI_API_KEY`。
 
 ---
 
@@ -84,7 +84,6 @@ npx @runapi.ai/mcp init roo
 | `create_task` | 是 | 创建媒体任务，并可选择轮询到完成。 |
 | `get_task` | 是 | 查询已有媒体任务的状态和最新 payload。 |
 | `check_balance` | 是 | 查询账户余额和消费指标。 |
-| `chat` | 是 | 向 RunAPI LLM 端点发送 messages，并在可用时返回 usage。 |
 
 ---
 
@@ -167,24 +166,10 @@ RunAPI 有哪些图片模型？
 1. 调用 `check_balance`。
 2. 如果没有配置 key，说明如何设置 `RUNAPI_API_KEY`。
 
-### LLM Chat
-
-```text
-用 RunAPI 的 LLM 模型总结这个文件。
-```
-
-预期行为：
-
-1. 必要时用目录工具查找当前 LLM model slug。
-2. 调用 `chat`，不要调用 `create_task`。
-3. 返回模型响应和可用的 usage metadata。
-
----
-
 ## Catalog Coverage
 
 内嵌目录来自 RunAPI 的 contract 快照。
-它包含媒体模型、工具类 endpoint 和 LLM model slugs。
+它包含媒体模型、工具类 endpoint 和用于发现的 LLM model slugs。
 
 | Modality | 使用方式 |
 |---|---|
@@ -196,6 +181,7 @@ RunAPI 有哪些图片模型？
 
 目录内容会随着版本变化。
 请用 `list_models` 获取当前 service/action/model slugs，用 `get_model_info` 查看当前约束。
+LLM 推理请直接通过 RunAPI API 或 SDK 接入。
 
 ---
 
diff --git a/evals/scenarios.md b/evals/scenarios.md
@@ -49,21 +49,14 @@ Use a real MCP host or the stdio client. Do not run credit-spending scenarios wi
 - Rule: Global behavior confirm before music; Phase 3 task creation.
 - Failure modes: spends credits without confirmation.
 
-### T3 — LLM Chat
-
-- Input: "Use a RunAPI LLM model to reply with exactly: RunAPI MCP check."
-- Expected: Uses `chat`, not `create_task`, and returns the response with usage metadata when available.
-- Rule: Phase 1 LLM chat.
-- Failure modes: calls media task endpoint, invents model slug.
-
-### T4 — Get Existing Task
+### T3 — Get Existing Task
 
 - Input: "Check task `<task-id>` for service `<service>` action `<action>`."
 - Expected: Calls `get_task` with service, action, and task_id.
 - Rule: Phase 1 checking.
 - Failure modes: omits action when known, creates a new task.
 
-### T5 — Submit Only
+### T4 — Submit Only
 
 - Input: "Create the task but do not wait for completion."
 - Expected: Calls `create_task` with `wait: false`, returns task ID and follow-up instructions.
diff --git a/glama.json b/glama.json
@@ -2,7 +2,7 @@
   "$schema": "https://glama.ai/mcp/schemas/server.json",
   "maintainers": ["runapi-builder"],
   "name": "RunAPI",
-  "description": "RunAPI MCP server for model discovery, pricing lookup, media task creation, polling, balance checks, and LLM chat.",
+  "description": "RunAPI MCP server for model discovery, pricing lookup, media task creation, polling, and balance checks.",
   "repository": "https://github.com/runapi-ai/mcp",
   "homepage": "https://runapi.ai",
   "license": "Apache-2.0",
diff --git a/package-lock.json b/package-lock.json
diff --git a/package.json b/package.json
@@ -1,7 +1,7 @@
 {
   "name": "@runapi.ai/mcp",
-  "version": "0.1.7",
-  "description": "RunAPI MCP server for model discovery, pricing lookup, task creation, and LLM chat.",
+  "version": "0.1.8",
+  "description": "RunAPI MCP server for model discovery, pricing lookup, media task creation, and balance checks.",
   "type": "module",
   "main": "dist/src/index.js",
   "bin": {
@@ -38,8 +38,7 @@
     "image-generation",
     "video-generation",
     "music-generation",
-    "audio-generation",
-    "llm"
+    "audio-generation"
   ],
   "author": "RunAPI",
   "license": "Apache-2.0",
diff --git a/plugin/.claude-plugin/plugin.json b/plugin/.claude-plugin/plugin.json
@@ -1,7 +1,7 @@
 {
   "name": "runapi",
-  "description": "RunAPI model discovery, pricing lookup, media task creation, polling, balance checks, and LLM chat for Claude Code.",
-  "version": "0.1.7",
+  "description": "RunAPI model discovery, pricing lookup, media task creation, polling, and balance checks for Claude Code.",
+  "version": "0.1.8",
   "author": {
     "name": "RunAPI",
     "url": "https://runapi.ai"
diff --git a/plugin/commands/setup.md b/plugin/commands/setup.md
@@ -11,7 +11,7 @@ Guide the user through local RunAPI MCP configuration.
 ## Instructions
 
 1. Explain that free catalog tools work without a key.
-2. Explain that `create_task`, `get_task`, `check_balance`, and `chat` require `RUNAPI_API_KEY`.
+2. Explain that `create_task`, `get_task`, and `check_balance` require `RUNAPI_API_KEY`.
 3. Tell the user to create a key in the RunAPI dashboard.
 4. Show one setup option:
 
diff --git a/plugin/hooks/check-config.sh b/plugin/hooks/check-config.sh
@@ -18,7 +18,7 @@ fi
 cat <<'EOF'
 RunAPI plugin loaded. Free catalog and prompt search tools are available now.
 
-To generate images, videos, music, or call LLMs:
+To generate images, videos, music, or audio:
   1. Sign up at https://runapi.ai and go to Dashboard > API Keys
   2. Save your key: mkdir -p ~/.config/runapi && echo '{"api_key":"YOUR_KEY"}' > ~/.config/runapi/config.json
   3. Restart Claude Code
diff --git a/server.json b/server.json
@@ -1,17 +1,17 @@
 {
   "$schema": "https://static.modelcontextprotocol.io/schemas/2025-12-11/server.schema.json",
   "name": "io.github.runapi-builder/runapi-mcp",
-  "description": "130+ AI models for image, video, music, audio, and LLM — 18 providers, one API key.",
+  "description": "130+ AI models for image, video, music, and audio — 18 providers, one API key.",
   "repository": {
     "url": "https://github.com/runapi-ai/mcp",
     "source": "github"
   },
-  "version": "0.1.5",
+  "version": "0.1.8",
   "packages": [
     {
       "registryType": "npm",
       "identifier": "@runapi.ai/mcp",
-      "version": "0.1.5",
+      "version": "0.1.8",
       "transport": {
         "type": "stdio"
       },
diff --git a/src/constants.ts b/src/constants.ts
@@ -1,5 +1,5 @@
 export const PACKAGE_NAME = "@runapi.ai/mcp";
-export const PACKAGE_VERSION = "0.1.7";
+export const PACKAGE_VERSION = "0.1.8";
 export const DEFAULT_BASE_URL = "https://runapi.ai";
 export const RUNAPI_API_KEY_ENV = "RUNAPI_API_KEY";
 export const USER_AGENT = `${PACKAGE_NAME}/${PACKAGE_VERSION}`;
diff --git a/src/lib/runapi-client.ts b/src/lib/runapi-client.ts
@@ -1,6 +1,6 @@
 import { USER_AGENT } from "../constants.js";
 import { loadConfig, requireApiKey, type RunApiConfig } from "../config.js";
-import type { ChatMessage, PollingOptions, RunApiPromptsResponse, RunApiTaskResponse, SearchPromptsParams } from "../types.js";
+import type { PollingOptions, RunApiPromptsResponse, RunApiTaskResponse, SearchPromptsParams } from "../types.js";
 import { errorFromResponse, PollTimeoutError } from "./errors.js";
 
 type RequestOptions = {
@@ -78,40 +78,6 @@ export class RunApiClient {
     throw new PollTimeoutError(`Timed out waiting for RunAPI task ${taskId}. Last status: ${taskStatus(lastTask)}.`);
   }
 
-  async chat(params: {
-    model: string;
-    messages: ChatMessage[];
-    max_tokens?: number;
-    temperature?: number;
-    api?: "messages" | "chat_completions";
-  }) {
-    if (params.api === "chat_completions") {
-      return this.request("POST", "/v1/chat/completions", {
-        auth: true,
-        body: {
-          model: params.model,
-          messages: params.messages,
-          max_tokens: params.max_tokens,
-          temperature: params.temperature
-        }
-      });
-    }
-
-    const system = params.messages.find((message) => message.role === "system")?.content;
-    const messages = params.messages.filter((message) => message.role !== "system");
-
-    return this.request("POST", "/v1/messages", {
-      auth: true,
-      body: {
-        model: params.model,
-        system,
-        messages,
-        max_tokens: params.max_tokens ?? 1024,
-        temperature: params.temperature
-      }
-    });
-  }
-
   private async request<T = unknown>(method: string, requestPath: string, options: RequestOptions = {}): Promise<T> {
     const url = new URL(requestPath, this.config.baseUrl.replace(/\/+$/, ""));
     const headers: Record<string, string> = {
diff --git a/src/server-instructions.ts b/src/server-instructions.ts
@@ -1,5 +1,5 @@
 export const SERVER_INSTRUCTIONS = `
-RunAPI gives this host model discovery, pricing lookup, task creation, task polling, balance checks, and LLM chat.
+RunAPI gives this host model discovery, pricing lookup, task creation, task polling, and balance checks.
 
 Available tools:
 - list_models: browse models by modality, service, or action. Free, no API key required.
@@ -10,7 +10,6 @@ Available tools:
 - check_balance: check account balance and spending. Requires API key.
 - create_task: create a media task and optionally poll for completion. Requires API key.
 - get_task: fetch current status and latest payload for an existing task. Requires API key.
-- chat: send messages to a RunAPI LLM endpoint. Requires API key.
 
 Global behavior:
 1. Use the user's language.
@@ -37,7 +36,7 @@ Phase 1: Intent assessment
 - If the user asks for prompt examples or inspiration, call search_prompts before drafting from scratch.
 - If the user asks to generate media, identify modality, service, action, model, and params from catalog tools.
 - If the user asks about an existing task, call get_task with service, action when known, and task_id.
-- If the user asks for LLM inference through RunAPI, use chat instead of create_task.
+- If the user asks for LLM inference through RunAPI, explain that this MCP server does not expose LLM inference tools and they should connect through the RunAPI API or SDK directly.
 
 Phase 2: Model selection
 - When the user names a model slug, use it after verifying with get_model_info.
diff --git a/src/tools/authenticated-handlers.ts b/src/tools/authenticated-handlers.ts
diff --git a/src/tools/authenticated.ts b/src/tools/authenticated.ts
diff --git a/src/types.ts b/src/types.ts
diff --git a/tests/integration/stdio-server.test.ts b/tests/integration/stdio-server.test.ts
diff --git a/tests/unit/authenticated-tools.test.ts b/tests/unit/authenticated-tools.test.ts
