mcp-duojie-image 0.9.0 → 0.10.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -5,12 +5,12 @@ MCP server for image generation via duojie.games GPT image API.
5
5
  ## Features
6
6
 
7
7
  - Text-to-image generation with `gpt-image-2`
8
- - Reference image support: requests with reference images route through `/v1/images/edits` (multipart form); text-only requests route through `/v1/images/generations` (JSON body)
8
+ - Two business operations: new-image generation with optional references, and explicit local editing
9
9
  - Prompt structuring helpers for use case, asset type, constraints, and effective prompt echo
10
10
  - Optional chroma-key transparent background post-processing
11
- - Async submission with task status lookup
11
+ - Async submission with explicit accepted/terminal semantics and required task polling
12
12
  - File saving with overwrite/conflict handling
13
- - Verified size templates for `1K`, `2K`, and `4K`
13
+ - Size request templates for `1K`, `2K`, and `4K`
14
14
 
15
15
  ## Configuration
16
16
 
@@ -23,7 +23,12 @@ Required:
23
23
 
24
24
  ## Tool: `generate_image`
25
25
 
26
- Generates an image asynchronously and saves it to disk.
26
+ Creates a new image. `reference_images` and `input_images` are optional guidance
27
+ for style, identity, composition, or material; they are never edit canvases.
28
+ A submitted response is not a successful generation. Call
29
+ `check_image_task` until it returns `status: "done"` or `status: "failed"`.
30
+ Non-terminal responses intentionally omit `success`; only terminal responses use
31
+ `success: true` or `success: false`.
27
32
 
28
33
  Required parameters:
29
34
 
@@ -39,7 +44,7 @@ Optional parameters:
39
44
  |---|---|---:|---|
40
45
  | `size` | `1K` \| `2K` \| `4K` | `1K` | Output size tier |
41
46
  | `aspect_ratio` | `1:1` \| `3:2` \| `2:3` \| `16:9` \| `9:16` | `1:1` | Verified ratio template |
42
- | `reference_images` | string[] | - | Legacy local reference image paths |
47
+ | `reference_images` | string[] | - | Local references used to create a new image |
43
48
  | `input_images` | object[] | - | Structured local image inputs with `path`, optional `role`, and optional `note` |
44
49
  | `use_case` | string | - | Prompt taxonomy slug, for example `product-mockup` or `background-extraction` |
45
50
  | `asset_type` | string | - | Where the asset will be used |
@@ -49,47 +54,80 @@ Optional parameters:
49
54
  | `format` | `png` \| `jpg` \| `jpeg` | `png` | Output format; content is encoded to match the extension |
50
55
  | `overwrite` | boolean | `false` | Overwrite existing file |
51
56
  | `negative_prompt` | string | - | Content to avoid |
52
- | `reference_max_side` | number | `1024` | Resize reference inputs to this max side before upload |
57
+ | `reference_max_side` | number | `512` / `256` | Default 512 for one generation reference and 256 for multiple |
53
58
  | `preserve_reference_detail` | boolean | `false` | Upload reference inputs without resizing |
54
59
  | `silent` | boolean | `false` | Disable desktop notification |
55
60
 
56
61
  Supported `input_images.role` values:
57
62
 
58
63
  - `reference`
59
- - `edit_target`
60
64
  - `style_reference`
61
65
  - `composition_reference`
62
66
  - `insert_source`
63
67
 
64
- **Endpoint routing:** Any request with reference images is sent to `/v1/images/edits` as a multipart form, which correctly handles both reference images and the `size`/`aspect_ratio` parameters. Text-only requests (no reference images) are sent to `/v1/images/generations` as a JSON body. The `role` field on `input_images` is used only for prompt structuring, not for endpoint selection.
68
+ | Tool | Meaning | Endpoint |
69
+ |---|---|---|
70
+ | `generate_image` | Create a new image; references are optional inputs | `/v1/images/generations` |
71
+ | `edit_image` | Modify one explicit target and preserve unspecified content | `/v1/images/edits` |
65
72
 
66
- When both `input_images` and `reference_images` are provided, `input_images` takes precedence. `generate_image` returns `effective_prompt` so callers can diagnose the exact prompt sent upstream.
73
+ When an image should only guide a new result, use `generate_image`. Use
74
+ `edit_image` only for an explicit local change such as removing an object or
75
+ changing one color while keeping the original. If the user intent is unclear,
76
+ ask before calling either tool. Supplying both image fields returns
77
+ `E_INPUT_MODE_CONFLICT`.
67
78
 
68
79
  Response:
69
80
 
70
81
  ```json
71
82
  {
72
- "success": true,
83
+ "accepted": true,
84
+ "terminal": false,
73
85
  "status": "submitted",
74
86
  "task_id": "uuid",
87
+ "poll_required": true,
88
+ "retry_after_ms": 1000,
89
+ "resolved_intent": "generate",
75
90
  "effective_prompt": "Primary request: ...",
76
- "message": "已提交异步生成;如需诊断失败原因,可调用 check_image_task"
91
+ "prompt_stats": {
92
+ "characters": 123,
93
+ "utf8Bytes": 145,
94
+ "lines": 4
95
+ },
96
+ "message": "已接受异步生成任务;这不代表图片已生成,必须轮询 check_image_task"
77
97
  }
78
98
  ```
79
99
 
100
+ ## Tool: `edit_image`
101
+
102
+ Edits exactly one original image. It shares the output, sizing, optional
103
+ supporting-reference, and asynchronous polling parameters above, and adds:
104
+
105
+ | Parameter | Type | Description |
106
+ |---|---|---|
107
+ | `edit_target` | string | Required local path of the single image to modify |
108
+
109
+ The edit target is always uploaded first. Supporting `reference_images` or
110
+ `input_images` follow it. The effective prompt tells the model to preserve every
111
+ unspecified detail.
112
+
113
+ Provider output is locally fitted to the requested size template before saving,
114
+ so upstream-native dimensions such as `1254x1254` or `1536x1024` do not leak
115
+ through the public contract.
116
+
80
117
  ## Tool: `check_image_task`
81
118
 
82
119
  Checks async task status and failure details.
83
120
 
84
121
  | Parameter | Type | Description |
85
122
  |---|---|---|
86
- | `task_id` | string | Task id returned by `generate_image` |
123
+ | `task_id` | string | Task id returned by `generate_image` or `edit_image` |
87
124
 
88
125
  Done response:
89
126
 
90
127
  ```json
91
128
  {
92
129
  "success": true,
130
+ "terminal": true,
93
131
  "task_id": "uuid",
94
132
  "status": "done",
95
133
  "saved_path": "D:/images/output.png",
@@ -102,16 +140,36 @@ Failed response:
102
140
  ```json
103
141
  {
104
142
  "success": false,
143
+ "terminal": true,
105
144
  "task_id": "uuid",
106
145
  "status": "failed",
107
- "error": "E_PROVIDER_BAD_REQUEST",
108
- "message": "Upstream error details"
146
+ "error": "E_PROVIDER_EMPTY_RESPONSE",
147
+ "message": "Provider returned HTTP 200 but data contained no image",
148
+ "details": {
149
+ "request_id": "provider-request-id",
150
+ "response_summary": {
151
+ "status": 200,
152
+ "content_type": "application/json",
153
+ "body": {
154
+ "kind": "json_object",
155
+ "keys": ["data"],
156
+ "data_length": 1,
157
+ "data_0_keys": []
158
+ }
159
+ }
160
+ }
109
161
  }
110
162
  ```
111
163
 
112
- ## Verified GPT size templates
164
+ Provider diagnostics contain only response metadata and JSON shape/key summaries;
165
+ prompt text, image payloads, authorization values, and arbitrary upstream body
166
+ content are not echoed. The configured provider timeout covers receiving headers,
167
+ reading the response body, and downloading a returned image URL.
168
+
169
+ ## GPT size request templates
113
170
 
114
- These templates were tested directly against `/v1/images/generations`; all returned the requested dimensions.
171
+ The server maps size tiers and ratios to these request dimensions. Actual output
172
+ dimensions still depend on the upstream provider response.
115
173
 
116
174
  | `size` | `1:1` | `3:2` | `2:3` | `16:9` | `9:16` |
117
175
  |---|---:|---:|---:|---:|---:|
package/dist/errors.js CHANGED
@@ -1,8 +1,10 @@
1
1
  export class AppError extends Error {
2
2
  code;
3
- constructor(code, message) {
3
+ details;
4
+ constructor(code, message, details) {
4
5
  super(message);
5
6
  this.code = code;
7
+ this.details = details;
6
8
  this.name = "AppError";
7
9
  }
8
10
  }
@@ -40,6 +40,32 @@ export async function encodeImage(imageBuffer, format) {
40
40
  throw new AppError("E_SAVE_FAILED", `Cannot encode image as ${format}: ${err}`);
41
41
  }
42
42
  }
43
+ export async function fitImageToSize(imageBuffer, targetSize) {
44
+ const match = targetSize.match(/^(\d+)x(\d+)$/);
45
+ if (!match) {
46
+ throw new AppError("E_INPUT_INVALID", `Invalid target image size: ${targetSize}`);
47
+ }
48
+ const width = Number(match[1]);
49
+ const height = Number(match[2]);
50
+ try {
51
+ const metadata = await sharp(imageBuffer, { failOn: "none" }).metadata();
52
+ if (metadata.width === width && metadata.height === height) {
53
+ return imageBuffer;
54
+ }
55
+ return await sharp(imageBuffer, { failOn: "none" })
56
+ .resize({
57
+ width,
58
+ height,
59
+ fit: "cover",
60
+ position: "attention",
61
+ })
62
+ .png()
63
+ .toBuffer();
64
+ }
65
+ catch (err) {
66
+ throw new AppError("E_SAVE_FAILED", `Cannot normalize image to ${targetSize}: ${err}`);
67
+ }
68
+ }
43
69
  export async function removeChromaKey(imageBuffer, keyColor = "#00ff00") {
44
70
  const key = parseHexColor(keyColor);
45
71
  const image = sharp(imageBuffer, { failOn: "none" }).ensureAlpha();
package/dist/index.js CHANGED
@@ -4,12 +4,13 @@ import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js"
4
4
  import { z } from "zod";
5
5
  import { createRequire } from "node:module";
6
6
  import { loadConfig } from "./config.js";
7
- import { ASPECT_RATIOS, IMAGE_SIZES, generateImage } from "./provider.js";
7
+ import { ASPECT_RATIOS, DEFAULT_GENERATION_REFERENCE_MAX_SIDE, DEFAULT_MULTI_REFERENCE_MAX_SIDE, IMAGE_SIZES, editImage, generateImage, mapOpenAISize, } from "./provider.js";
8
8
  import { saveImage } from "./saver.js";
9
9
  import { AppError } from "./errors.js";
10
10
  import { createTask, getTask, updateTask } from "./task-queue.js";
11
- import { IMAGE_INPUT_ROLES, PROMPT_USE_CASES, preparePrompt } from "./prompt.js";
12
- import { OUTPUT_FORMATS, removeChromaKey } from "./image-processing.js";
11
+ import { taskStatusResponse, taskSubmissionResponse } from "./task-response.js";
12
+ import { PROMPT_USE_CASES, REFERENCE_IMAGE_ROLES, preparePrompt, } from "./prompt.js";
13
+ import { OUTPUT_FORMATS, fitImageToSize, removeChromaKey, } from "./image-processing.js";
13
14
  const require = createRequire(import.meta.url);
14
15
  // eslint-disable-next-line @typescript-eslint/no-explicit-any
15
16
  const notifier = require("node-notifier");
@@ -29,66 +30,43 @@ function toolText(data) {
29
30
  }],
30
31
  };
31
32
  }
33
+ function toolError(code, message) {
34
+ return {
35
+ ...toolText({ success: false, error: code, message }),
36
+ isError: true,
37
+ };
38
+ }
39
+ function validateTransparentOutput(args) {
40
+ if (!args.transparent_background)
41
+ return undefined;
42
+ const filenameExt = args.output_filename.match(/\.([^.]+)$/)?.[1]?.toLowerCase();
43
+ if ((args.format && args.format !== "png") || (filenameExt && filenameExt !== "png")) {
44
+ return toolError("E_INPUT_INVALID", "transparent_background requires PNG output");
45
+ }
46
+ return undefined;
47
+ }
32
48
  const server = new McpServer({
33
49
  name: "duojie-image",
34
- version: "0.1.0",
50
+ version: "0.10.0",
35
51
  });
36
- server.registerTool("generate_image", {
37
- description: "Generate image async with gpt-image-2. Reference images route through /v1/images/edits (multipart form); text-only requests route through /v1/images/generations. Return after tool call; don't wait/check.",
38
- inputSchema: z.object({
39
- prompt: z.string().min(1).describe("Prompt"),
40
- output_dir: z.string().describe("Absolute output dir"),
41
- output_filename: z.string().describe("Filename without ext"),
42
- reference_images: z.array(z.string()).optional().describe("Reference image paths"),
43
- input_images: z.array(z.object({
44
- path: z.string().min(1).describe("Local image path"),
45
- role: z.enum(IMAGE_INPUT_ROLES).optional().describe("How this image should guide generation/editing"),
46
- note: z.string().optional().describe("Short role-specific instruction"),
47
- })).optional().describe("Structured image inputs with roles; supersedes reference_images when provided"),
48
- use_case: z.enum(PROMPT_USE_CASES).optional().describe("Prompt taxonomy slug"),
49
- asset_type: z.string().optional().describe("Where the image will be used"),
50
- constraints: z.string().optional().describe("Must-keep or must-avoid constraints"),
51
- transparent_background: z.boolean().optional().describe("Generate on a chroma-key background and remove it locally; PNG only"),
52
- transparent_key_color: z.string().regex(/^#[0-9a-fA-F]{6}$/).optional().describe("Chroma-key color, default #00ff00"),
53
- size: z.enum(IMAGE_SIZES).optional().describe("Default 1K. Supported: 1K, 2K, 4K"),
54
- aspect_ratio: z.enum(ASPECT_RATIOS).optional().describe("Supported: 1:1, 3:2, 2:3, 16:9, 9:16"),
55
- format: z.enum(OUTPUT_FORMATS).optional().describe("Default png. Supported: png, jpg, jpeg"),
56
- overwrite: z.boolean().optional().describe("Default false"),
57
- negative_prompt: z.string().optional().describe("Avoid"),
58
- reference_max_side: z.number().int().min(256).max(3840).optional().describe("Resize reference inputs to this max side; default 1024"),
59
- preserve_reference_detail: z.boolean().optional().describe("Do not resize reference inputs before upload"),
60
- silent: z.boolean().optional().describe("No notification"),
61
- }),
62
- }, async (args) => {
52
+ async function submitImageTask(args, operation) {
63
53
  let config;
64
54
  try {
65
55
  config = loadConfig();
66
56
  }
67
57
  catch (err) {
68
58
  const code = err instanceof AppError ? err.code : "E_CONFIG_MISSING_KEY";
69
- return { content: [{ type: "text", text: JSON.stringify({ success: false, error: code, message: String(err instanceof Error ? err.message : err) }) }], isError: true };
70
- }
71
- const silent = args.silent ?? false;
72
- if (args.transparent_background) {
73
- const filenameExt = args.output_filename.match(/\.([^.]+)$/)?.[1]?.toLowerCase();
74
- if ((args.format && args.format !== "png") || (filenameExt && filenameExt !== "png")) {
75
- return {
76
- content: [{
77
- type: "text",
78
- text: JSON.stringify({
79
- success: false,
80
- error: "E_INPUT_INVALID",
81
- message: "transparent_background requires PNG output",
82
- }),
83
- }],
84
- isError: true,
85
- };
86
- }
59
+ return toolError(code, String(err instanceof Error ? err.message : err));
87
60
  }
61
+ const transparentError = validateTransparentOutput(args);
62
+ if (transparentError)
63
+ return transparentError;
88
64
  let preparedPrompt;
89
65
  try {
90
66
  preparedPrompt = preparePrompt({
67
+ operation,
91
68
  prompt: args.prompt,
69
+ editTarget: args.edit_target,
92
70
  negativePrompt: args.negative_prompt,
93
71
  useCase: args.use_case,
94
72
  assetType: args.asset_type,
@@ -101,21 +79,37 @@ server.registerTool("generate_image", {
101
79
  }
102
80
  catch (err) {
103
81
  const code = err instanceof AppError ? err.code : "E_INPUT_INVALID";
104
- return { content: [{ type: "text", text: JSON.stringify({ success: false, error: code, message: String(err instanceof Error ? err.message : err) }) }], isError: true };
82
+ return toolError(code, String(err instanceof Error ? err.message : err));
105
83
  }
106
84
  const task = createTask();
107
- updateTask(task.id, { effectivePrompt: preparedPrompt.prompt });
108
- // Start background generation (don't await)
109
- (async () => {
85
+ updateTask(task.id, {
86
+ effectivePrompt: preparedPrompt.prompt,
87
+ promptStats: preparedPrompt.stats,
88
+ resolvedIntent: preparedPrompt.resolvedIntent,
89
+ });
90
+ void (async () => {
110
91
  try {
111
92
  updateTask(task.id, { status: "running", startedAt: Date.now() });
112
- let imageBuffer = await generateImage(config, {
113
- prompt: preparedPrompt.prompt,
114
- referenceImages: preparedPrompt.imagePaths,
115
- size: args.size ?? "1K",
116
- aspectRatio: args.aspect_ratio,
117
- referenceMaxSide: args.preserve_reference_detail ? false : args.reference_max_side,
118
- });
93
+ const referenceMaxSide = args.preserve_reference_detail
94
+ ? false
95
+ : args.reference_max_side;
96
+ let imageBuffer = operation === "edit"
97
+ ? await editImage(config, {
98
+ prompt: preparedPrompt.prompt,
99
+ editTarget: preparedPrompt.imagePaths[0],
100
+ referenceImages: preparedPrompt.imagePaths.slice(1),
101
+ size: args.size ?? "1K",
102
+ aspectRatio: args.aspect_ratio,
103
+ referenceMaxSide,
104
+ })
105
+ : await generateImage(config, {
106
+ prompt: preparedPrompt.prompt,
107
+ referenceImages: preparedPrompt.imagePaths,
108
+ size: args.size ?? "1K",
109
+ aspectRatio: args.aspect_ratio,
110
+ referenceMaxSide,
111
+ });
112
+ imageBuffer = await fitImageToSize(imageBuffer, mapOpenAISize(args.size ?? "1K", args.aspect_ratio));
119
113
  const outputFormat = args.transparent_background ? "png" : args.format;
120
114
  if (args.transparent_background) {
121
115
  imageBuffer = await removeChromaKey(imageBuffer, args.transparent_key_color);
@@ -133,10 +127,10 @@ server.registerTool("generate_image", {
133
127
  });
134
128
  await server.sendLoggingMessage({
135
129
  level: "info",
136
- data: `图片生成完成 [${task.id}]:${savedPath}`,
130
+ data: `图片${operation === "edit" ? "编辑" : "生成"}完成 [${task.id}]:${savedPath}`,
137
131
  });
138
- if (!silent)
139
- notify("图片生成完成", savedPath);
132
+ if (!args.silent)
133
+ notify("图片任务完成", savedPath);
140
134
  }
141
135
  catch (err) {
142
136
  const code = err instanceof AppError ? err.code : "E_INTERNAL";
@@ -146,67 +140,74 @@ server.registerTool("generate_image", {
146
140
  completedAt: Date.now(),
147
141
  errorCode: code,
148
142
  error: msg,
143
+ errorDetails: err instanceof AppError ? err.details : undefined,
149
144
  });
150
145
  await server.sendLoggingMessage({
151
146
  level: "error",
152
- data: `图片生成失败 [${task.id}] [${code}]: ${msg}`,
147
+ data: `图片任务失败 [${task.id}] [${code}]: ${msg}`,
153
148
  }).catch(() => { });
154
- if (!silent) {
155
- if (code === "E_PROVIDER_TIMEOUT") {
156
- notify("图片生成超时", `${args.output_filename} 生成超过 300 秒,请重试`);
157
- }
158
- else {
159
- notify("图片生成失败", `[${code}] ${msg}`);
160
- }
161
- }
149
+ if (!args.silent)
150
+ notify("图片任务失败", `[${code}] ${msg}`);
162
151
  }
163
152
  })();
164
- return {
165
- content: [{
166
- type: "text",
167
- text: JSON.stringify({
168
- success: true,
169
- status: "submitted",
170
- task_id: task.id,
171
- effective_prompt: preparedPrompt.prompt,
172
- message: `已提交异步生成:${args.output_filename};如需诊断失败原因,可调用 check_image_task`,
173
- }),
174
- }],
175
- };
176
- });
153
+ return toolText(taskSubmissionResponse(getTask(task.id) ?? task, args.output_filename));
154
+ }
155
+ const structuredReferences = z.array(z.object({
156
+ path: z.string().min(1).describe("Local reference image path"),
157
+ role: z.enum(REFERENCE_IMAGE_ROLES).optional().describe("reference | style_reference | composition_reference | insert_source"),
158
+ note: z.string().optional().describe("Short role-specific instruction"),
159
+ })).optional().describe("Structured references; do not combine with reference_images");
160
+ const commonFields = {
161
+ prompt: z.string().min(1).describe("Image request"),
162
+ output_dir: z.string().describe("Absolute output directory"),
163
+ output_filename: z.string().describe("Filename without extension"),
164
+ use_case: z.enum(PROMPT_USE_CASES).optional().describe("Prompt taxonomy slug"),
165
+ asset_type: z.string().optional().describe("Where the image will be used"),
166
+ constraints: z.string().optional().describe("Must-keep or must-avoid constraints"),
167
+ size: z.enum(IMAGE_SIZES).optional().describe("Default 1K"),
168
+ aspect_ratio: z.enum(ASPECT_RATIOS).optional().describe("1:1 | 3:2 | 2:3 | 16:9 | 9:16"),
169
+ format: z.enum(OUTPUT_FORMATS).optional().describe("Default png"),
170
+ overwrite: z.boolean().optional().describe("Default false"),
171
+ negative_prompt: z.string().optional().describe("Content to avoid"),
172
+ reference_max_side: z.number().int().min(256).max(3840).optional()
173
+ .describe(`Input resize limit. generate defaults: ${DEFAULT_GENERATION_REFERENCE_MAX_SIDE} for one reference, ${DEFAULT_MULTI_REFERENCE_MAX_SIDE} for multiple; edit default 1024`),
174
+ preserve_reference_detail: z.boolean().optional().describe("Upload inputs without resizing; less stable"),
175
+ silent: z.boolean().optional().describe("Disable desktop notification"),
176
+ };
177
+ const referenceFields = {
178
+ reference_images: z.array(z.string()).optional()
179
+ .describe("Optional local reference paths for creating a new image; never treated as edit canvases"),
180
+ input_images: structuredReferences,
181
+ };
182
+ server.registerTool("generate_image", {
183
+ description: "Create a NEW image from a prompt, optionally guided by reference images. This is the default for style, identity, composition, material, scene-change, and same-style requests; references are never edit canvases. Do not use for a local change that must preserve the original. If the user intent is ambiguous, ask whether they want a new referenced image or a small edit before calling. Async: poll check_image_task until done or failed.",
184
+ inputSchema: z.object({
185
+ ...commonFields,
186
+ ...referenceFields,
187
+ transparent_background: z.boolean().optional().describe("Generate on chroma key and remove it locally; PNG only"),
188
+ transparent_key_color: z.string().regex(/^#[0-9a-fA-F]{6}$/).optional().describe("Default #00ff00"),
189
+ }),
190
+ }, async (args) => submitImageTask(args, "generate"));
191
+ server.registerTool("edit_image", {
192
+ description: "Make a SMALL, explicit change to one existing image while preserving everything not requested. Use only when the user clearly asks to modify the original (for example remove an object, change a color, or replace a local element). If it is unclear whether they want a new image or an edit, ask before calling. Exactly one edit_target is required. Async: poll check_image_task until done or failed.",
193
+ inputSchema: z.object({
194
+ ...commonFields,
195
+ edit_target: z.string().min(1).describe("The single original image to modify"),
196
+ ...referenceFields,
197
+ }),
198
+ }, async (args) => submitImageTask(args, "edit"));
177
199
  server.registerTool("check_image_task", {
178
- description: "Check an async image generation task status and failure details.",
200
+ description: "Poll a generate_image or edit_image task. Continue while poll_required=true. Only status=done with success=true means an image was saved; failed is terminal with diagnostics.",
179
201
  inputSchema: z.object({
180
- task_id: z.string().min(1).describe("Task id returned by generate_image"),
202
+ task_id: z.string().min(1).describe("Task id returned by an image tool"),
181
203
  }),
182
204
  }, async (args) => {
183
205
  const task = getTask(args.task_id);
184
206
  if (!task) {
185
- return {
186
- ...toolText({
187
- success: false,
188
- error: "E_TASK_NOT_FOUND",
189
- message: `Task ${args.task_id} not found. It may have expired or never existed.`,
190
- }),
191
- isError: true,
192
- };
193
- }
194
- const response = {
195
- success: task.status !== "failed",
196
- task_id: task.id,
197
- status: task.status,
198
- };
199
- if (task.status === "done") {
200
- response.saved_path = task.savedPath;
201
- }
202
- if (task.effectivePrompt) {
203
- response.effective_prompt = task.effectivePrompt;
204
- }
205
- if (task.status === "failed") {
206
- response.error = task.errorCode;
207
- response.message = task.error;
207
+ return toolError("E_TASK_NOT_FOUND", `Task ${args.task_id} not found. It may have expired or never existed.`);
208
208
  }
209
- return toolText(response);
209
+ const response = toolText(taskStatusResponse(task));
210
+ return task.status === "failed" ? { ...response, isError: true } : response;
210
211
  });
211
212
  const transport = new StdioServerTransport();
212
213
  await server.connect(transport);
package/dist/prompt.js CHANGED
@@ -1,7 +1,6 @@
1
1
  import { AppError } from "./errors.js";
2
- export const IMAGE_INPUT_ROLES = [
2
+ export const REFERENCE_IMAGE_ROLES = [
3
3
  "reference",
4
- "edit_target",
5
4
  "style_reference",
6
5
  "composition_reference",
7
6
  "insert_source",
@@ -34,12 +33,28 @@ function compactLines(lines) {
34
33
  function validateImageInputs(inputImages) {
35
34
  if (!inputImages)
36
35
  return undefined;
37
- for (const [index, image] of inputImages.entries()) {
36
+ return inputImages.map((image, index) => {
38
37
  if (!image.path?.trim()) {
39
38
  throw new AppError("E_INPUT_INVALID", `input_images[${index}].path is required`);
40
39
  }
40
+ return { ...image, path: image.path.trim() };
41
+ });
42
+ }
43
+ function legacyReferenceInputs(referenceImages) {
44
+ if (!referenceImages?.length)
45
+ return undefined;
46
+ return referenceImages.map((path, index) => {
47
+ if (!path?.trim()) {
48
+ throw new AppError("E_INPUT_INVALID", `reference_images[${index}] is required`);
49
+ }
50
+ return { path: path.trim(), role: "reference" };
51
+ });
52
+ }
53
+ function resolveReferenceInputs(opts) {
54
+ if (opts.inputImages !== undefined && opts.referenceImages !== undefined) {
55
+ throw new AppError("E_INPUT_MODE_CONFLICT", "Use either input_images or reference_images, not both");
41
56
  }
42
- return inputImages;
57
+ return validateImageInputs(opts.inputImages) ?? legacyReferenceInputs(opts.referenceImages);
43
58
  }
44
59
  function imageRoleLabel(role) {
45
60
  return role?.replace(/_/g, " ") ?? "reference";
@@ -48,15 +63,18 @@ function imageLines(inputImages) {
48
63
  if (!inputImages?.length)
49
64
  return undefined;
50
65
  const lines = inputImages.map((image, index) => {
51
- const note = image.note?.trim() ? `; ${image.note.trim()}` : "";
66
+ const note = "note" in image && image.note?.trim() ? `; ${image.note.trim()}` : "";
52
67
  return `Image ${index + 1}: ${imageRoleLabel(image.role)}${note}`;
53
68
  });
54
69
  return `Input images: ${lines.join("; ")}`;
55
70
  }
56
- function legacyReferenceInputs(referenceImages) {
57
- if (!referenceImages?.length)
58
- return undefined;
59
- return referenceImages.map((path) => ({ path, role: "reference" }));
71
+ function operationPrompt(operation, hasReferences) {
72
+ if (operation === "edit") {
73
+ return "Intent: make only the requested changes to Image 1; preserve every unspecified object, identity, composition, camera angle, material, text, lighting, and detail. Later images are supporting references only.";
74
+ }
75
+ return hasReferences
76
+ ? "Intent: create a new image; use input images only as style, composition, identity, or material references. Do not edit or reproduce any source canvas."
77
+ : "Intent: create a new image from the request.";
60
78
  }
61
79
  function transparentPrompt(keyColor) {
62
80
  return compactLines([
@@ -66,7 +84,23 @@ function transparentPrompt(keyColor) {
66
84
  ]);
67
85
  }
68
86
  export function preparePrompt(opts) {
69
- const inputImages = validateImageInputs(opts.inputImages) ?? legacyReferenceInputs(opts.referenceImages);
87
+ const references = resolveReferenceInputs(opts);
88
+ let imageInputs;
89
+ if (opts.operation === "edit") {
90
+ if (!opts.editTarget?.trim()) {
91
+ throw new AppError("E_INPUT_MODE_CONFLICT", "edit_image requires one edit_target");
92
+ }
93
+ imageInputs = [
94
+ { path: opts.editTarget.trim(), role: "edit_target" },
95
+ ...(references ?? []),
96
+ ];
97
+ }
98
+ else {
99
+ if (opts.editTarget !== undefined) {
100
+ throw new AppError("E_INPUT_MODE_CONFLICT", "generate_image does not accept edit_target; use edit_image for local changes");
101
+ }
102
+ imageInputs = references;
103
+ }
70
104
  const keyColor = opts.transparentKeyColor ?? DEFAULT_KEY_COLOR;
71
105
  const negativePrompt = opts.negativePrompt?.trim()
72
106
  ? `Avoid: ${opts.negativePrompt.trim()}`
@@ -74,7 +108,8 @@ export function preparePrompt(opts) {
74
108
  const prompt = compactLines([
75
109
  opts.useCase ? `Use case: ${opts.useCase}` : undefined,
76
110
  opts.assetType ? `Asset type: ${opts.assetType}` : undefined,
77
- imageLines(inputImages),
111
+ operationPrompt(opts.operation, Boolean(references?.length)),
112
+ imageLines(imageInputs),
78
113
  `Primary request: ${opts.prompt.trim()}`,
79
114
  opts.constraints ? `Constraints: ${opts.constraints.trim()}` : undefined,
80
115
  opts.transparentBackground ? transparentPrompt(keyColor) : undefined,
@@ -82,7 +117,13 @@ export function preparePrompt(opts) {
82
117
  ]);
83
118
  return {
84
119
  prompt,
85
- imagePaths: inputImages?.map((image) => image.path) ?? [],
86
- imageInputs: inputImages,
120
+ resolvedIntent: opts.operation,
121
+ imagePaths: imageInputs?.map((image) => image.path) ?? [],
122
+ imageInputs,
123
+ stats: {
124
+ characters: Array.from(prompt).length,
125
+ utf8Bytes: Buffer.byteLength(prompt, "utf8"),
126
+ lines: prompt ? prompt.split("\n").length : 0,
127
+ },
87
128
  };
88
129
  }
package/dist/provider.js CHANGED
@@ -2,42 +2,102 @@ import { AppError } from "./errors.js";
2
2
  import { resizeIfNeeded } from "./resizer.js";
3
3
  export const ASPECT_RATIOS = ["1:1", "3:2", "2:3", "16:9", "9:16"];
4
4
  export const IMAGE_SIZES = ["1K", "2K", "4K"];
5
- async function fetchWithRetry(url, init, timeoutMs, retries = 0) {
5
+ function getRequestId(res) {
6
+ return res.headers.get("x-request-id")
7
+ ?? res.headers.get("request-id")
8
+ ?? res.headers.get("x-trace-id")
9
+ ?? res.headers.get("cf-ray");
10
+ }
11
+ function summarizeJson(value) {
12
+ if (Array.isArray(value)) {
13
+ return { kind: "json_array", length: value.length };
14
+ }
15
+ if (!value || typeof value !== "object") {
16
+ return { kind: "json_scalar", type: value === null ? "null" : typeof value };
17
+ }
18
+ const record = value;
19
+ const summary = {
20
+ kind: "json_object",
21
+ keys: Object.keys(record).slice(0, 20),
22
+ };
23
+ if (Array.isArray(record.data)) {
24
+ summary.data_length = record.data.length;
25
+ const first = record.data[0];
26
+ if (first && typeof first === "object" && !Array.isArray(first)) {
27
+ summary.data_0_keys = Object.keys(first).slice(0, 20);
28
+ }
29
+ }
30
+ if (record.error && typeof record.error === "object" && !Array.isArray(record.error)) {
31
+ const error = record.error;
32
+ summary.error_keys = Object.keys(error).slice(0, 20);
33
+ if (typeof error.code === "string" && /^[a-zA-Z0-9_.:-]{1,80}$/.test(error.code)) {
34
+ summary.error_code = error.code;
35
+ }
36
+ if (typeof error.type === "string" && /^[a-zA-Z0-9_.:-]{1,80}$/.test(error.type)) {
37
+ summary.error_type = error.type;
38
+ }
39
+ }
40
+ return summary;
41
+ }
42
+ function responseDetails(res, body) {
43
+ const summary = {
44
+ status: res.status,
45
+ content_type: res.headers.get("content-type"),
46
+ content_length: res.headers.get("content-length"),
47
+ ...(body ? { body } : {}),
48
+ };
49
+ return {
50
+ request_id: getRequestId(res),
51
+ response_summary: summary,
52
+ };
53
+ }
54
+ function isAbortError(err) {
55
+ return err instanceof Error && (err.name === "AbortError" || err.name === "TimeoutError");
56
+ }
57
+ async function delayForRetry(attempt) {
58
+ const delayMs = 1_000 * Math.pow(2, attempt) + Math.random() * 500;
59
+ await new Promise((resolve) => setTimeout(resolve, delayMs));
60
+ }
61
+ async function fetchWithRetry(url, init, timeoutMs, consume, retries = 0) {
6
62
  let lastErr;
7
63
  for (let attempt = 0; attempt <= retries; attempt++) {
8
64
  const controller = new AbortController();
9
65
  const timer = setTimeout(() => controller.abort(), timeoutMs);
66
+ let receivedResponse;
10
67
  try {
11
68
  const res = await fetch(url, { ...init, signal: controller.signal });
12
- clearTimeout(timer);
69
+ receivedResponse = res;
13
70
  // 5xx: 上游随机故障(EOF等),立刻 abort body 连接并重试,不等 chunked 终止
14
71
  if (res.status >= 500 && attempt < retries) {
15
- try {
16
- controller.abort();
17
- }
18
- catch { }
19
- const delayMs = 1000 * Math.pow(2, attempt) + Math.random() * 500;
20
- await new Promise((r) => setTimeout(r, delayMs));
72
+ controller.abort();
73
+ clearTimeout(timer);
74
+ await delayForRetry(attempt);
21
75
  continue;
22
76
  }
23
- return res;
77
+ const result = await consume(res, controller.signal);
78
+ clearTimeout(timer);
79
+ return result;
24
80
  }
25
81
  catch (err) {
26
82
  clearTimeout(timer);
27
- const isAbort = err instanceof Error && err.name === "AbortError";
28
- // AbortError from our manual abort above is not a real timeout
29
- if (isAbort && attempt < retries) {
30
- lastErr = err;
31
- const delayMs = 1000 * Math.pow(2, attempt) + Math.random() * 500;
32
- await new Promise((r) => setTimeout(r, delayMs));
33
- continue;
83
+ if (err instanceof AppError) {
84
+ throw err;
85
+ }
86
+ if (isAbortError(err)) {
87
+ throw new AppError("E_PROVIDER_TIMEOUT", `Provider request timed out after ${timeoutMs}ms`, {
88
+ timeout_ms: timeoutMs,
89
+ phase: receivedResponse ? "response_body_or_image_download" : "connection_or_headers",
90
+ ...(receivedResponse ? responseDetails(receivedResponse) : {}),
91
+ });
34
92
  }
35
- if (isAbort) {
36
- throw new AppError("E_PROVIDER_TIMEOUT", "Request timed out");
93
+ // Once response headers have arrived, the generation may already be billable.
94
+ // Do not replay the POST for body/CDN failures and risk duplicate generation.
95
+ if (receivedResponse) {
96
+ throw new AppError("E_PROVIDER_BAD_REQUEST", "Provider response could not be read completely", responseDetails(receivedResponse));
37
97
  }
38
98
  lastErr = err;
39
99
  if (attempt < retries) {
40
- await new Promise((r) => setTimeout(r, 1000 * Math.pow(2, attempt)));
100
+ await delayForRetry(attempt);
41
101
  }
42
102
  }
43
103
  }
@@ -67,90 +127,130 @@ const SIZE_TEMPLATES = {
67
127
  "9:16": "2160x3840",
68
128
  },
69
129
  };
70
- function mapOpenAISize(size, aspectRatio) {
130
+ export function mapOpenAISize(size, aspectRatio) {
71
131
  const imageSize = size ?? "1K";
72
132
  const ratio = aspectRatio ?? "1:1";
73
133
  return SIZE_TEMPLATES[imageSize][ratio];
74
134
  }
75
- async function imageFromOpenAIResponse(res) {
76
- const data = (await res.json());
135
+ async function imageFromOpenAIResponse(res, signal) {
136
+ let data;
137
+ try {
138
+ data = (await res.json());
139
+ }
140
+ catch (err) {
141
+ if (isAbortError(err))
142
+ throw err;
143
+ if (!(err instanceof SyntaxError)) {
144
+ throw new AppError("E_PROVIDER_BAD_REQUEST", "Provider response body could not be read", responseDetails(res));
145
+ }
146
+ throw new AppError("E_PROVIDER_EMPTY_RESPONSE", "Provider returned HTTP 200 but the response body was not valid image JSON", responseDetails(res, { kind: "invalid_json" }));
147
+ }
77
148
  const image = data.data?.[0];
78
149
  if (image?.b64_json) {
79
- return Buffer.from(image.b64_json, "base64");
150
+ const imageBuffer = Buffer.from(image.b64_json, "base64");
151
+ if (imageBuffer.length > 0) {
152
+ return imageBuffer;
153
+ }
80
154
  }
81
155
  if (image?.url) {
82
- const imgRes = await fetch(image.url);
83
- if (!imgRes.ok) {
84
- throw new AppError("E_PROVIDER_BAD_REQUEST", `Image URL fetch failed: ${imgRes.status}`);
156
+ try {
157
+ const imgRes = await fetch(image.url, { signal });
158
+ if (!imgRes.ok) {
159
+ const summary = await readBodySummary(imgRes);
160
+ throw new AppError("E_PROVIDER_BAD_REQUEST", `Image URL fetch failed (HTTP ${imgRes.status})`, responseDetails(imgRes, summary));
161
+ }
162
+ const imageBuffer = Buffer.from(await imgRes.arrayBuffer());
163
+ if (imageBuffer.length > 0) {
164
+ return imageBuffer;
165
+ }
166
+ throw new AppError("E_PROVIDER_EMPTY_RESPONSE", "Provider image URL returned an empty body", responseDetails(imgRes, { kind: "empty", length: 0 }));
167
+ }
168
+ catch (err) {
169
+ if (isAbortError(err) || err instanceof AppError)
170
+ throw err;
171
+ throw new AppError("E_PROVIDER_BAD_REQUEST", "Image URL download failed", responseDetails(res));
85
172
  }
86
- return Buffer.from(await imgRes.arrayBuffer());
87
173
  }
88
- throw new AppError("E_PROVIDER_BAD_REQUEST", "No image in response");
174
+ throw new AppError("E_PROVIDER_EMPTY_RESPONSE", "Provider returned HTTP 200 but data contained no image", responseDetails(res, summarizeJson(data)));
175
+ }
176
+ async function readBodySummary(res) {
177
+ const text = await res.text();
178
+ if (!text)
179
+ return { kind: "empty", length: 0 };
180
+ try {
181
+ return summarizeJson(JSON.parse(text));
182
+ }
183
+ catch {
184
+ return { kind: "text", length: text.length };
185
+ }
89
186
  }
90
187
  async function handleProviderError(res) {
188
+ // 5xx: 上游随机故障,不读 body(chunked 连接可能挂起),直接返回安全的响应头摘要。
189
+ if (res.status >= 500) {
190
+ throw new AppError("E_PROVIDER_BAD_REQUEST", `Upstream error (HTTP ${res.status})`, responseDetails(res));
191
+ }
192
+ const summary = await readBodySummary(res);
193
+ const details = responseDetails(res, summary);
91
194
  if (res.status === 401 || res.status === 403) {
92
- throw new AppError("E_PROVIDER_AUTH", `Auth failed (HTTP ${res.status})`);
195
+ throw new AppError("E_PROVIDER_AUTH", `Auth failed (HTTP ${res.status})`, details);
93
196
  }
94
197
  if (res.status === 429) {
95
- throw new AppError("E_PROVIDER_RATE_LIMIT", "Rate limit exceeded");
198
+ throw new AppError("E_PROVIDER_RATE_LIMIT", "Rate limit exceeded", details);
96
199
  }
97
- // 5xx: 上游随机故障,不读 body(chunked 连接挂 120s),直接秒回
98
- if (res.status >= 500) {
99
- throw new AppError("E_PROVIDER_BAD_REQUEST", `Upstream error ${res.status} (body skipped to avoid chunked hang)`);
100
- }
101
- const text = await res.text().catch(() => "");
102
- throw new AppError("E_PROVIDER_BAD_REQUEST", `Upstream error ${res.status}: ${text.slice(0, 200)}`);
200
+ throw new AppError("E_PROVIDER_BAD_REQUEST", `Upstream error (HTTP ${res.status})`, details);
103
201
  }
104
- async function generateOpenAIImage(config, opts) {
105
- const prompt = opts.prompt;
106
- const size = mapOpenAISize(opts.size, opts.aspectRatio);
107
- const hasReferenceImages = Boolean(opts.referenceImages?.length);
108
- // 有参考图 → /v1/images/edits(multipart form)
109
- // 无参考图 → /v1/images/generations(JSON body,纯文本)
110
- const useEdits = hasReferenceImages;
111
- const url = `${config.baseUrl}${useEdits ? "/v1/images/edits" : "/v1/images/generations"}`;
112
- let body;
113
- let headers = {
114
- Authorization: `Bearer ${config.apiKey}`,
115
- };
116
- if (useEdits) {
117
- // 有参考图:multipart form
118
- const form = new FormData();
119
- form.set("model", GPT_IMAGE_MODEL);
120
- form.set("prompt", prompt);
121
- form.set("size", size);
122
- if (opts.aspectRatio) {
123
- form.set("aspect_ratio", opts.aspectRatio);
124
- }
125
- for (const [index, imgPath] of opts.referenceImages.entries()) {
126
- const data = await resizeIfNeeded(imgPath, opts.referenceMaxSide);
127
- const bytes = data.buffer.slice(data.byteOffset, data.byteOffset + data.byteLength);
128
- form.append("image", new Blob([bytes], { type: "image/webp" }), `reference-${index}.webp`);
129
- }
130
- body = form;
131
- }
132
- else {
133
- // 纯文本生成
134
- headers = { ...headers, "Content-Type": "application/json" };
135
- body = JSON.stringify({
136
- model: GPT_IMAGE_MODEL,
137
- prompt,
138
- size,
139
- ...(opts.aspectRatio ? { aspect_ratio: opts.aspectRatio } : {}),
140
- });
141
- }
142
- // 启用 2 次重试,应对上游偶发的 EOF / 524(Cloudflare 100s 超时)。
143
- // retries=0 时所有偶发失败一次过即死,指数退避代码是死代码。
144
- const res = await fetchWithRetry(url, {
202
+ async function postForImage(config, path, body, headers, parseImage) {
203
+ return fetchWithRetry(`${config.baseUrl}${path}`, {
145
204
  method: "POST",
146
205
  headers,
147
206
  body,
148
- }, config.timeoutMs, 2);
149
- if (!res.ok) {
150
- await handleProviderError(res);
207
+ }, config.timeoutMs, async (res, signal) => {
208
+ if (!res.ok) {
209
+ await handleProviderError(res);
210
+ }
211
+ return parseImage(res, signal);
212
+ }, 2);
213
+ }
214
+ export const DEFAULT_GENERATION_REFERENCE_MAX_SIDE = 512;
215
+ export const DEFAULT_MULTI_REFERENCE_MAX_SIDE = 256;
216
+ async function generationBody(opts) {
217
+ const imageUrls = [];
218
+ const referenceImages = opts.referenceImages ?? [];
219
+ const defaultMaxSide = referenceImages.length > 1
220
+ ? DEFAULT_MULTI_REFERENCE_MAX_SIDE
221
+ : DEFAULT_GENERATION_REFERENCE_MAX_SIDE;
222
+ for (const imgPath of referenceImages) {
223
+ const maxSide = opts.referenceMaxSide === undefined
224
+ ? defaultMaxSide
225
+ : opts.referenceMaxSide;
226
+ const data = await resizeIfNeeded(imgPath, maxSide);
227
+ imageUrls.push(`data:image/webp;base64,${data.toString("base64")}`);
151
228
  }
152
- return imageFromOpenAIResponse(res);
229
+ return JSON.stringify({
230
+ model: GPT_IMAGE_MODEL,
231
+ prompt: opts.prompt,
232
+ size: mapOpenAISize(opts.size, opts.aspectRatio),
233
+ ...(imageUrls.length ? { image: imageUrls } : {}),
234
+ ...(opts.aspectRatio ? { aspect_ratio: opts.aspectRatio } : {}),
235
+ });
236
+ }
237
+ export async function editImage(config, opts) {
238
+ const imagePaths = [opts.editTarget, ...(opts.referenceImages ?? [])];
239
+ const form = new FormData();
240
+ form.set("model", GPT_IMAGE_MODEL);
241
+ form.set("prompt", opts.prompt);
242
+ form.set("size", mapOpenAISize(opts.size, opts.aspectRatio));
243
+ const fieldName = imagePaths.length > 1 ? "image[]" : "image";
244
+ for (const [index, imgPath] of imagePaths.entries()) {
245
+ const data = await resizeIfNeeded(imgPath, opts.referenceMaxSide);
246
+ const bytes = data.buffer.slice(data.byteOffset, data.byteOffset + data.byteLength);
247
+ form.append(fieldName, new Blob([bytes], { type: "image/webp" }), `input-${index + 1}.webp`);
248
+ }
249
+ return postForImage(config, "/v1/images/edits", form, { Authorization: `Bearer ${config.apiKey}` }, imageFromOpenAIResponse);
153
250
  }
154
251
  export async function generateImage(config, opts) {
155
- return generateOpenAIImage(config, opts);
252
+ return postForImage(config, "/v1/images/generations", await generationBody(opts), {
253
+ Authorization: `Bearer ${config.apiKey}`,
254
+ "Content-Type": "application/json",
255
+ }, imageFromOpenAIResponse);
156
256
  }
package/dist/resizer.js CHANGED
@@ -4,7 +4,7 @@ import { AppError } from "./errors.js";
4
4
  export const DEFAULT_REFERENCE_MAX_SIDE = 1024;
5
5
  /**
6
6
  * 读入垫图,按需缩放到 maxSide 内(保持比例不放大),
7
- * 强制重编码为 WebP(quality 85)以满足 gpt-image-2 edits 端点的 1.5MB 甜点值约束。
7
+ * 强制重编码为 WebP(quality 85),降低多模态与编辑请求的上传体积。
8
8
  * PNG 通道比 WebP 大 3~5 倍,改用 WebP 后体积大幅下降,避免上游 100s 超时。
9
9
  */
10
10
  export async function resizeIfNeeded(imgPath, maxSide = DEFAULT_REFERENCE_MAX_SIDE) {
@@ -0,0 +1,49 @@
1
+ const POLL_RETRY_AFTER_MS = 1_000;
2
+ export function taskSubmissionResponse(task, outputFilename) {
3
+ return {
4
+ accepted: true,
5
+ terminal: false,
6
+ status: "submitted",
7
+ task_id: task.id,
8
+ poll_required: true,
9
+ retry_after_ms: POLL_RETRY_AFTER_MS,
10
+ resolved_intent: task.resolvedIntent,
11
+ effective_prompt: task.effectivePrompt,
12
+ prompt_stats: task.promptStats,
13
+ message: `已接受异步生成任务:${outputFilename}。这不代表图片已生成;必须调用 check_image_task,直到 status 为 done 或 failed。`,
14
+ };
15
+ }
16
+ export function taskStatusResponse(task) {
17
+ const response = {
18
+ task_id: task.id,
19
+ status: task.status,
20
+ terminal: task.status === "done" || task.status === "failed",
21
+ };
22
+ if (task.effectivePrompt) {
23
+ response.effective_prompt = task.effectivePrompt;
24
+ }
25
+ if (task.promptStats) {
26
+ response.prompt_stats = task.promptStats;
27
+ }
28
+ if (task.resolvedIntent) {
29
+ response.resolved_intent = task.resolvedIntent;
30
+ }
31
+ if (task.status === "done") {
32
+ response.success = true;
33
+ response.saved_path = task.savedPath;
34
+ }
35
+ else if (task.status === "failed") {
36
+ response.success = false;
37
+ response.error = task.errorCode;
38
+ response.message = task.error;
39
+ if (task.errorDetails) {
40
+ response.details = task.errorDetails;
41
+ }
42
+ }
43
+ else {
44
+ response.accepted = true;
45
+ response.poll_required = true;
46
+ response.retry_after_ms = POLL_RETRY_AFTER_MS;
47
+ }
48
+ return response;
49
+ }
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "mcp-duojie-image",
3
- "version": "0.9.0",
3
+ "version": "0.10.0",
4
4
  "description": "MCP server for image generation via duojie.games",
5
5
  "type": "module",
6
6
  "main": "dist/index.js",