paddle-ocr-server 0.1.0__py3-none-any.whl

paddle_ocr_server-0.1.0.dist-info/METADATA

@@ -0,0 +1,92 @@
+Metadata-Version: 2.4
+Name: paddle-ocr-server
+Version: 0.1.0
+Summary: MCP server for PaddleOCR PDF-to-Markdown conversion
+Requires-Python: >=3.13
+Requires-Dist: fastmcp>=3.4.2
+Requires-Dist: httpx>=0.28.0
+Description-Content-Type: text/markdown
+
+# PaddleOCR MCP Server
+
+通过 PaddleOCR 官方 API 将 PDF 或图片转换为 Markdown。
+
+## 功能
+
+- **Tools**: `ocr_pdf` — 提交 PDF/图片到 PaddleOCR，返回合并后的 Markdown 文件路径
+
+## 环境要求
+
+- Python >= 3.13
+- [uv](https://docs.astral.sh/uv/)
+
+## 环境变量
+
+| 变量 | 必需 | 说明 |
+|------|------|------|
+| `PADDLEOCR_ACCESS_TOKEN` | ✅ | AI Studio access token，从 https://aistudio.baidu.com/account/accessToken 获取 |
+
+## 安装
+
+```bash
+cd servers/paddle-ocr
+uv sync
+```
+
+## 运行
+
+```bash
+PADDLEOCR_ACCESS_TOKEN=your-token uv run python server.py
+```
+
+## 测试
+
+```bash
+uv run pytest -v
+```
+
+## 接入 MCP Host
+
+发布到 PyPI 后可用 `uvx`：
+
+```json
+{
+    "mcpServers": {
+        "paddle-ocr": {
+            "command": "uvx",
+            "args": ["paddle-ocr-server"],
+            "env": {
+                "PADDLEOCR_ACCESS_TOKEN": "your-token"
+            }
+        }
+    }
+}
+```
+
+本地开发时用 `uv`：
+
+```json
+{
+    "mcpServers": {
+        "paddle-ocr": {
+            "command": "uv",
+            "args": ["--directory", "/path/to/servers/paddle-ocr", "run", "paddle-ocr-server"],
+            "env": {
+                "PADDLEOCR_ACCESS_TOKEN": "your-token"
+            }
+        }
+    }
+}
+```
+```
+
+## 使用方式
+
+调用 `ocr_pdf` tool，提供 `file_path`（本地文件）或 `file_url`（URL），二选一。
+
+结果保存到当前工作目录下的同名文件夹，结构如下：
+- `full.md` — 合并所有页面的 Markdown
+- `pages/page_N.md` — 每页单独的 Markdown
+- `imgs/` — 图片文件
+
+返回值为 `full.md` 的绝对路径。

paddle_ocr_server-0.1.0.dist-info/RECORD

@@ -0,0 +1,5 @@
+server.py,sha256=Gyj-OyZtsx0K2TlLehJV6W1asZuTMeUl8T_eqD6TSjU,6964
+paddle_ocr_server-0.1.0.dist-info/METADATA,sha256=cAbwIDguFDUjyBEiG_4_nif4Uef_0F90ObfKP_xG1H4,1862
+paddle_ocr_server-0.1.0.dist-info/WHEEL,sha256=mffPy8wBnZQn2VnJUU5jE99KsxaSfiyMHV9Yt0aLVxs,87
+paddle_ocr_server-0.1.0.dist-info/entry_points.txt,sha256=SbUtF8LurvLyiCcMBe4Xg4Iotm3AX-SKpLNLXrh-Ds8,50
+paddle_ocr_server-0.1.0.dist-info/RECORD,,

paddle_ocr_server-0.1.0.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: hatchling 1.30.1
+Root-Is-Purelib: true
+Tag: py3-none-any

paddle_ocr_server-0.1.0.dist-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+paddle-ocr-server = server:main

server.py

@@ -0,0 +1,220 @@
+"""PaddleOCR MCP Server — PDF/图片 → Markdown via PaddleOCR 官方 API.
+
+暴露一个 tool: ocr_pdf
+  - 输入本地文件路径或 URL
+  - 提交异步 OCR 任务，轮询直到完成
+  - 下载结果，保存 Markdown 和图片到当前工作目录下的同名文件夹（可指定 output_dir 覆盖）
+  - 返回合并后的 Markdown 文件路径
+
+环境变量:
+  PADDLEOCR_ACCESS_TOKEN  — 必需，AI Studio access token
+"""
+
+import asyncio
+import json
+import os
+import re
+import time
+from pathlib import Path
+
+import httpx
+from fastmcp import FastMCP
+
+mcp = FastMCP("PaddleOCR")
+
+API_BASE = "https://paddleocr.aistudio-app.com/api/v2/ocr/jobs"
+DEFAULT_MODEL = "PaddleOCR-VL-1.6"
+POLL_INTERVAL = 5
+POLL_TIMEOUT = 600
+
+
+def _get_token() -> str:
+    token = os.environ.get("PADDLEOCR_ACCESS_TOKEN", "")
+    if not token:
+        raise ValueError("PADDLEOCR_ACCESS_TOKEN 环境变量未设置")
+    return token
+
+
+def _auth_headers(token: str) -> dict[str, str]:
+    return {"Authorization": f"bearer {token}"}
+
+
+async def _submit_job(
+    file_path: str | None,
+    file_url: str | None,
+    use_chart_recognition: bool = True,
+) -> str:
+    """提交 OCR 任务，返回 jobId。"""
+    token = _get_token()
+    headers = _auth_headers(token)
+
+    optional_payload = {
+        "useDocOrientationClassify": False,
+        "useDocUnwarping": False,
+        "useChartRecognition": use_chart_recognition,
+    }
+
+    async with httpx.AsyncClient(timeout=120) as client:
+        if file_url:
+            headers["Content-Type"] = "application/json"
+            payload = {
+                "fileUrl": file_url,
+                "model": DEFAULT_MODEL,
+                "optionalPayload": optional_payload,
+            }
+            resp = await client.post(API_BASE, json=payload, headers=headers)
+        else:
+            data = {
+                "model": DEFAULT_MODEL,
+                "optionalPayload": json.dumps(optional_payload),
+            }
+            files = {"file": open(file_path, "rb")}  # noqa: SIM115
+            resp = await client.post(API_BASE, headers=headers, data=data, files=files)
+
+    resp.raise_for_status()
+    return resp.json()["data"]["jobId"]
+
+
+async def _poll_until_done(job_id: str) -> str:
+    """轮询任务状态直到完成，返回 jsonUrl。"""
+    token = _get_token()
+    headers = _auth_headers(token)
+    deadline = time.monotonic() + POLL_TIMEOUT
+
+    async with httpx.AsyncClient(timeout=60) as client:
+        while True:
+            resp = await client.get(f"{API_BASE}/{job_id}", headers=headers)
+            resp.raise_for_status()
+            body = resp.json()["data"]
+            state = body["state"]
+
+            if state == "done":
+                return body["resultUrl"]["jsonUrl"]
+            if state == "failed":
+                raise RuntimeError(f"OCR 任务失败: {body.get('errorMsg', '未知错误')}")
+
+            if time.monotonic() > deadline:
+                raise TimeoutError(f"OCR 任务轮询超时 ({POLL_TIMEOUT}s)")
+
+            await asyncio.sleep(POLL_INTERVAL)
+
+
+def _re_process(text: str) -> str:
+    """Markdown 后处理：修复 LaTeX 公式、压缩空行。"""
+    # 去掉行内公式的空格: $ ... $ → $...$
+    text = re.sub(r"(?<!\$)\$\s*([^$]+?)\s*\$(?!\$)", r"$\1$", text)
+    # 块公式标准化: $$...$$ 前后加空行
+    text = re.sub(
+        r"\$\$([\s\S]*?)\$\$",
+        lambda m: f"\n$$\n{m.group(1).strip()}\n$$\n",
+        text,
+    )
+    # 压缩多余空行
+    text = re.sub(r"\n{3,}", "\n\n", text)
+    return text
+
+
+async def _download_and_save(jsonl_url: str, output_dir: Path) -> Path:
+    """下载 JSONL 结果，保存 Markdown 和图片到 output_dir，返回合并后的 md 路径。"""
+    async with httpx.AsyncClient(timeout=120) as client:
+        resp = await client.get(jsonl_url)
+        resp.raise_for_status()
+
+    output_dir.mkdir(parents=True, exist_ok=True)
+    pages_dir = output_dir / "pages"
+    pages_dir.mkdir(parents=True, exist_ok=True)
+    pages_md: list[str] = []
+    page_num = 0
+
+    for line in resp.text.strip().split("\n"):
+        line = line.strip()
+        if not line:
+            continue
+        result = json.loads(line)["result"]
+
+        for res in result["layoutParsingResults"]:
+            md_text = _re_process(res["markdown"]["text"])
+            pages_md.append(md_text)
+
+            # 保存单页 md
+            page_file = pages_dir / f"page_{page_num}.md"
+            page_file.write_text(md_text, encoding="utf-8")
+
+            # 下载 markdown 中引用的图片
+            for img_path, img_url in res["markdown"].get("images", {}).items():
+                full_img_path = output_dir / img_path
+                full_img_path.parent.mkdir(parents=True, exist_ok=True)
+                img_bytes = await _fetch_bytes(img_url)
+                full_img_path.write_bytes(img_bytes)
+
+            page_num += 1
+
+    # 合并所有页面到单个 md
+    merged = output_dir / "full.md"
+    merged.write_text("\n\n---\n\n".join(pages_md), encoding="utf-8")
+    return merged
+
+
+async def _fetch_bytes(url: str) -> bytes:
+    async with httpx.AsyncClient(timeout=60) as client:
+        resp = await client.get(url)
+        resp.raise_for_status()
+        return resp.content
+
+
+@mcp.tool
+async def ocr_pdf(
+    file_path: str | None = None,
+    file_url: str | None = None,
+    output_dir: str | None = None,
+    use_chart_recognition: bool = True,
+) -> str:
+    """将 PDF 或图片文件通过 PaddleOCR 转换为 Markdown。
+
+    提供 file_path（本地文件）或 file_url（URL），二选一。
+    结果保存到指定目录，返回合并后的 Markdown 文件路径。
+
+    Args:
+        file_path: 本地文件的绝对路径（PDF/图片）。
+        file_url: 文件的 URL 地址。
+        output_dir: 输出目录的绝对路径。未指定时默认保存到当前工作目录下的同名文件夹。
+        use_chart_recognition: 是否启用图表识别，默认 True。
+    """
+    if not file_path and not file_url:
+        raise ValueError("必须提供 file_path 或 file_url 其中之一")
+    if file_path and file_url:
+        raise ValueError("file_path 和 file_url 只能提供一个")
+
+    src = None
+    if file_path:
+        src = Path(file_path)
+        if not src.exists():
+            raise FileNotFoundError(f"文件不存在: {file_path}")
+
+    # 确定输出目录
+    if output_dir:
+        out = Path(output_dir)
+    elif src:
+        out = Path.cwd() / src.stem
+    else:
+        from urllib.parse import urlparse
+
+        url_path = urlparse(file_url).path
+        stem = Path(url_path).stem or "ocr_output"
+        out = Path.cwd() / stem
+
+    # 提交 → 轮询 → 取结果
+    job_id = await _submit_job(file_path, file_url, use_chart_recognition)
+    jsonl_url = await _poll_until_done(job_id)
+    merged_md = await _download_and_save(jsonl_url, out)
+
+    return str(merged_md)
+
+
+
+def main():
+    mcp.run()
+
+
+if __name__ == "__main__":
+    main()