Develop

.dockerignore

@@ -0,0 +1,19 @@
+.git
+.github
+.hypothesis
+.mypy_cache
+.pytest_cache
+.ruff_cache
+.venv
+.aci
+.compare
+__pycache__
+dist
+build
+tests
+doc
+*.pyc
+*.pyo
+*.pyd
+*.log
+uv.lock

Dockerfile

@@ -0,0 +1,16 @@
+FROM python:3.12-slim
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1
+
+WORKDIR /build
+
+COPY pyproject.toml README.md ./
+COPY src ./src
+
+RUN pip install --no-cache-dir uv \
+    && uv pip install --system .
+
+WORKDIR /data
+
+ENTRYPOINT ["aci-mcp"]

README.md

@@ -83,6 +83,7 @@ aci shell
 ```
 
 This launches an interactive REPL (Read-Eval-Print Loop) with:
+
 - Command history (up/down arrows to navigate)
 - Tab completion for commands
 - Persistent history across sessions
@@ -102,7 +103,7 @@ This launches an interactive REPL (Read-Eval-Print Loop) with:
 
 ### Example Session
 
-```
+```text
 $ aci shell
 
     _    ____ ___   ____  _          _ _ 
@@ -142,6 +143,7 @@ Search queries support inline modifiers to filter results:
 | `exclude:<pattern>` | Alias for `-path:` | `exclude:fixtures` |
 
 Multiple exclusions can be combined:
+
 ```bash
 aci search "database query -path:tests -path:fixtures"
 ```
@@ -190,13 +192,53 @@ ACI supports the Model Context Protocol (MCP), allowing LLMs to directly interac
 }
 ```
 
-2. Ensure `.env` exists in the working directory with required settings (see `.env.example`)
+1. Ensure `.env` exists in the working directory with required settings (see `.env.example`)
 
-3. Use natural language to interact with your codebase:
+2. Use natural language to interact with your codebase:
    - "Index the current directory"
    - "Search for authentication functions"
    - "Show me the index status"
 
+### Docker Sidecar Delivery
+
+For agentic coding tools, the recommended deployment model is a local Docker sidecar:
+
+- The code repository stays on the user's machine
+- The MCP server runs in a local container
+- Qdrant runs either as another local container or as a cloud endpoint
+- The embedding API uses the user's own API key
+
+Build the image:
+
+```bash
+docker build -t aci-mcp:latest .
+```
+
+If you want a local Qdrant container, start it separately:
+
+```bash
+docker run -d --name aci-qdrant -p 6333:6333 qdrant/qdrant:latest
+```
+
+Then configure your MCP client to launch ACI through Docker. A complete template is available in `mcp-config.docker.example.json`.
+
+Important runtime rules:
+
+- Mount the host source tree read-only into the container, for example `/workspace`
+- Persist `/data` as a Docker volume so `.aci/index.db` survives container restarts
+- Set `ACI_MCP_WORKSPACE_ROOT` for relative paths
+- Set `ACI_MCP_PATH_MAPPINGS` when the MCP client sends host-native absolute paths such as `D:\repo` or `/Users/alice/repo`
+
+Example mapping values:
+
+```text
+ACI_MCP_WORKSPACE_ROOT=/workspace
+ACI_MCP_PATH_MAPPINGS=D:\repo=/workspace
+ACI_MCP_PATH_MAPPINGS=/Users/alice/repo=/workspace
+```
+
+When path mappings are configured, MCP tools can accept the host path provided by the client and resolve it to the mounted container path automatically.
+
 ### Available MCP Tools
 
 | Tool | Description |
@@ -235,6 +277,7 @@ REINDEX=1 uv run python scripts/measure_mcp_search.py
 ### Debug Mode
 
 Set `ACI_ENV=development` in `.env` to enable debug logging:
+
 ```
 ACI_ENV=development
 ```
@@ -266,6 +309,7 @@ cp .env.example .env
 ```
 
 Key settings:
+
 | Variable | Description | Required |
 |----------|-------------|----------|
 | `ACI_EMBEDDING_API_KEY` | API key for embedding service | Yes |
@@ -275,6 +319,8 @@ Key settings:
 | `ACI_VECTOR_STORE_API_KEY` | Qdrant API key (for Qdrant Cloud) | No |
 | `ACI_VECTOR_STORE_HOST` | Qdrant host | No (defaults to localhost) |
 | `ACI_VECTOR_STORE_PORT` | Qdrant port | No (defaults to 6333) |
+| `ACI_MCP_WORKSPACE_ROOT` | Base directory for relative MCP paths inside the container/runtime | No |
+| `ACI_MCP_PATH_MAPPINGS` | Host-to-container path prefix mappings for MCP, separated by `;` | No |
 | `ACI_SERVER_HOST` | HTTP server host | No (defaults to 0.0.0.0) |
 | `ACI_SERVER_PORT` | HTTP server port | No (defaults to 8000) |
 | `ACI_ENV` | Environment (development/production) | No |
@@ -284,3 +330,6 @@ See `.env.example` for the full list of options.
 The CLI and HTTP server will attempt to auto-start a local Qdrant Docker container only when
 targeting a local endpoint (`localhost` / `127.0.0.1`). For cloud Qdrant (`ACI_VECTOR_STORE_URL`),
 it will not run Docker.
+
+When ACI itself is running inside a container, it will not attempt to launch nested Docker for Qdrant.
+In that setup, run Qdrant as a separate local container or point `ACI_VECTOR_STORE_URL` to Qdrant Cloud.

doc/README.zh-CN.md

@@ -82,6 +82,7 @@ aci shell
 ```
 
 启动后会进入 REPL（Read-Eval-Print Loop），包含：
+
 - 命令历史（方向键上下浏览）
 - 命令自动补全（Tab）
 - 跨会话持久化历史
@@ -190,13 +191,53 @@ ACI 支持 Model Context Protocol（MCP），使 LLM 可直接调用你的代码
 }
 ```
 
-2. 确保工作目录存在 `.env` 且配置完整（参考 `../.env.example`）
+1. 确保工作目录存在 `.env` 且配置完整（参考 `../.env.example`）
 
-3. 可用自然语言与代码库交互，例如：
+2. 可用自然语言与代码库交互，例如：
    - “索引当前目录”
    - “搜索认证相关函数”
    - “查看当前索引状态”
 
+### 以 Docker Sidecar 交付 MCP
+
+对于 Agentic coding tools，推荐的交付模型是本地 Docker sidecar：
+
+- 代码仓库保留在用户机器上
+- MCP server 运行在本地容器里
+- Qdrant 可以是另一个本地容器，也可以是云端地址
+- Embedding API 使用用户自己的 key
+
+构建镜像：
+
+```bash
+docker build -t aci-mcp:latest .
+```
+
+如果使用本地 Qdrant 容器，单独启动它：
+
+```bash
+docker run -d --name aci-qdrant -p 6333:6333 qdrant/qdrant:latest
+```
+
+然后让 MCP 客户端通过 Docker 拉起 ACI。完整模板见 `mcp-config.docker.example.json`。
+
+运行时约定：
+
+- 把宿主机源码目录以只读方式挂载进容器，例如 `/workspace`
+- 把 `/data` 挂成 Docker volume，这样 `.aci/index.db` 不会随容器重建丢失
+- 相对路径场景设置 `ACI_MCP_WORKSPACE_ROOT`
+- 如果 MCP 客户端传的是宿主机绝对路径，例如 `D:\repo` 或 `/Users/alice/repo`，设置 `ACI_MCP_PATH_MAPPINGS`
+
+示例：
+
+```text
+ACI_MCP_WORKSPACE_ROOT=/workspace
+ACI_MCP_PATH_MAPPINGS=D:\repo=/workspace
+ACI_MCP_PATH_MAPPINGS=/Users/alice/repo=/workspace
+```
+
+配置后，MCP 工具可以接受客户端传来的宿主机路径，并自动解析到容器内的挂载路径。
+
 ### MCP 可用工具
 
 | 工具 | 说明 |
@@ -277,10 +318,14 @@ cp .env.example .env
 | `ACI_VECTOR_STORE_API_KEY` | Qdrant API Key（Qdrant Cloud） | 否 |
 | `ACI_VECTOR_STORE_HOST` | Qdrant 主机地址 | 否（默认 localhost） |
 | `ACI_VECTOR_STORE_PORT` | Qdrant 端口 | 否（默认 6333） |
+| `ACI_MCP_WORKSPACE_ROOT` | MCP 在容器/运行时中解析相对路径时使用的基础目录 | 否 |
+| `ACI_MCP_PATH_MAPPINGS` | MCP 使用的宿主路径前缀到容器路径前缀映射，使用 `;` 分隔 | 否 |
 | `ACI_SERVER_HOST` | HTTP 服务主机地址 | 否（默认 0.0.0.0） |
 | `ACI_SERVER_PORT` | HTTP 服务端口 | 否（默认 8000） |
 | `ACI_ENV` | 运行环境（development/production） | 否 |
 
 完整配置请查看 `../.env.example`。
 
 CLI 和 HTTP 服务仅在目标为本地端点（`localhost` / `127.0.0.1`）时尝试自动启动本地 Qdrant Docker 容器。若使用云端 Qdrant（`ACI_VECTOR_STORE_URL`），不会启动 Docker。
+
+当 ACI 自身运行在容器内时，不会再尝试嵌套启动 Docker 来拉起 Qdrant。此时请把 Qdrant 作为独立本地容器运行，或直接配置 `ACI_VECTOR_STORE_URL` 指向 Qdrant Cloud。

mcp-config.docker.example.json

@@ -0,0 +1,39 @@
+{
+  "mcpServers": {
+    "aci": {
+      "command": "docker",
+      "args": [
+        "run",
+        "-i",
+        "--rm",
+        "-v",
+        "HOST_SOURCE_PATH:/workspace:ro",
+        "-v",
+        "aci-mcp-data:/data",
+        "-e",
+        "ACI_MCP_WORKSPACE_ROOT=/workspace",
+        "-e",
+        "ACI_MCP_PATH_MAPPINGS=HOST_SOURCE_PATH=/workspace",
+        "-e",
+        "ACI_EMBEDDING_API_URL",
+        "-e",
+        "ACI_EMBEDDING_API_KEY",
+        "-e",
+        "ACI_EMBEDDING_MODEL",
+        "-e",
+        "ACI_EMBEDDING_DIMENSION",
+        "-e",
+        "ACI_VECTOR_STORE_URL=http://host.docker.internal:6333",
+        "-e",
+        "ACI_VECTOR_STORE_VECTOR_SIZE=1024",
+        "aci-mcp:latest"
+      ],
+      "env": {
+        "ACI_EMBEDDING_API_URL": "https://api.openai.com/v1/embeddings",
+        "ACI_EMBEDDING_API_KEY": "your_api_key_here",
+        "ACI_EMBEDDING_MODEL": "text-embedding-3-small",
+        "ACI_EMBEDDING_DIMENSION": "1024"
+      }
+    }
+  }
+}

pyproject.toml

@@ -22,7 +22,7 @@ classifiers = [
     "Topic :: Utilities",
 ]
 
-# 运行时核心依赖（字母顺序）
+# runtime dependencies
 dependencies = [
     "fastapi>=0.111.0",
     "httpx>=0.25.0",

src/aci/core/__init__.py

@@ -41,11 +41,13 @@
     ScannedFile,
     get_default_registry,
 )
-from aci.core.tokenizer import (
-    TiktokenTokenizer,
-    TokenizerInterface,
-    get_default_tokenizer,
-)
+from aci.core.tokenizer import (
+    CharacterTokenizer,
+    SimpleTokenizer,
+    TiktokenTokenizer,
+    TokenizerInterface,
+    get_default_tokenizer,
+)
 from aci.core.watch_config import WatchConfig
 
 __all__ = [
@@ -70,10 +72,12 @@
     "TreeSitterParser",
     "SUPPORTED_LANGUAGES",
     "check_tree_sitter_setup",
-    # Tokenizer
-    "TokenizerInterface",
-    "TiktokenTokenizer",
-    "get_default_tokenizer",
+    # Tokenizer
+    "TokenizerInterface",
+    "TiktokenTokenizer",
+    "CharacterTokenizer",
+    "SimpleTokenizer",
+    "get_default_tokenizer",
     # Chunker
     "CodeChunk",
     "ChunkerConfig",

src/aci/core/config.py

@@ -133,6 +133,7 @@ class IndexingConfig:
         default_factory=lambda: _get_default("indexing", "chunk_overlap_lines", 2)
     )
     max_workers: int = field(default_factory=lambda: _get_default("indexing", "max_workers", 4))
+    tokenizer: str = field(default_factory=lambda: _get_default("indexing", "tokenizer", "tiktoken"))
 
 
 @dataclass
@@ -226,6 +227,7 @@ def apply_env_overrides(self) -> "ACIConfig":
             "ACI_INDEXING_MAX_CHUNK_TOKENS": ("indexing", "max_chunk_tokens", int),
             "ACI_INDEXING_CHUNK_OVERLAP_LINES": ("indexing", "chunk_overlap_lines", int),
             "ACI_INDEXING_MAX_WORKERS": ("indexing", "max_workers", int),
+            "ACI_TOKENIZER": ("indexing", "tokenizer", str),
             "ACI_INDEXING_FILE_EXTENSIONS": (
                 "indexing",
                 "file_extensions",