ContextMCP
Dockerized MCP server for context management with vector search, supporting local (Ollama + Pinecone Local) and cloud modes, multi-repo indexing, and backup/restore workflows.
README
ContextMCP Dockerized Setup
This repository wraps the forked J3ys/context-mcp project and provides two Docker Compose entrypoints:
docker-compose.local.ymlfor fully local development with Ollama + Pinecone Localdocker-compose.cloud.ymlfor app/reindex flows backed by cloud services
The application logic now lives in the fork, while this repository keeps the Docker, config, and helper-script layer around it.
Local mode
- Copy the example environment file:
cp .env.example .env
-
Review
config.yamland replace the placeholder source before indexing. -
Start Pinecone Local and Ollama:
docker compose -f docker-compose.local.yml up -d pinecone ollama
- Pull the default embedding model:
docker compose -f docker-compose.local.yml --profile bootstrap run --rm ollama-bootstrap
Alternatively, if Ollama is running directly on the host:
ollama pull nomic-embed-text
- Run reindex:
docker compose -f docker-compose.local.yml --profile reindex run --rm reindex
- Start the app:
docker compose -f docker-compose.local.yml up -d app
Local Pinecone backup and restore
Pinecone Local is in-memory and does not provide native backup/restore. This repository adds a repo-owned export/import workflow on top of the local API.
Because the local index is in memory, it is a good idea to create a backup right after a successful indexing run. That way you can restore the same local corpus after a reboot instead of rerunning a long reindex.
Before using the backup/restore scripts on a fresh clone, install the small host-side helper dependency:
npm install
Create a backup:
./scripts/backup-local-index.sh
This writes a new folder under ./backups/, for example:
backups/contextmcp-docs-2026-07-31T12-00-00Z/
Restore from the latest backup:
./scripts/restore-local-index.sh
Restore from a specific backup:
BACKUP_DIR=./backups/contextmcp-docs-2026-07-31T12-00-00Z ./scripts/restore-local-index.sh
Useful knobs:
BACKUP_ROOT- where backup folders are createdBACKUP_DIR- explicitly restore from a specific backup folderBACKUP_BATCH_SIZE- vectors per exported backup fileRESTORE_INDEX_NAME- restore into a different local index nameRESTORE_CLEAR_FIRST=false- keep existing vectors and upsert on topRESTORE_UPSERT_BATCH_SIZE- vectors per restore upsert request
Restore behavior notes:
- the restore script will create the target index if it does not exist
- if the target index already exists, restore validates dimension and metric compatibility before clearing anything
Local multi-repo indexing (optional)
Instead of editing config.yaml per repository, you can index an arbitrary
list of local repositories by setting two variables in .env:
LOCAL_DOCS_ROOT=/absolute/path/to/your/projects
LOCAL_DOCS_REPOS=repo-one=java,repo-two=typescript:sub/path
LOCAL_DOCS_ROOTis mounted read-only into thereindexcontainer at/workspace/repos.LOCAL_DOCS_REPOSis a comma-separated list ofrepo=languagepairs. Eachrepomust be a subdirectory name underLOCAL_DOCS_ROOT.languagemust be one ofjava,typescript, orjavascript. You can optionally append a repo-relative subpath after the language (repo=language:sub/path) when the actual source code lives below the repo root, for exampleHybris=java:hybris/bin/custom.- Each repo is indexed as one mixed source set: the AST-aware
parser: codechunker extracts method/function/class chunks from code files and whole-file chunks from text-like config/infra files (for example yaml/yml/json/ properties/xml/toml/conf/cfg/ini, shell scripts, Dockerfiles, and common dot-config files such as.gitignore,.dockerignore,.editorconfig). Common build/tooling directories are still skipped (.git,node_modules,build,.gradle,.idea, etc.). - When
LOCAL_DOCS_REPOSis unset, this feature does nothing and reindex falls back to the source configured inconfig.yaml, unchanged.
Run it the same way as the standard local flow:
docker compose -f docker-compose.local.yml --profile reindex run --rm reindex
docker compose -f docker-compose.local.yml up -d app
Cloud mode
- Copy the example environment file:
cp .env.example .env
- Change
.envfor cloud usage:
- set
PINECONE_MODE=cloud - set
PINECONE_API_KEYto your real key - clear
PINECONE_CONTROLLER_HOST - set
OPENAI_API_KEY, or updateconfig.cloud.yamlto a different embedding provider before setting that provider's key
-
Review
config.cloud.yamland replace the placeholder source before indexing. -
Run the cloud reindex/app flow:
docker compose -f docker-compose.cloud.yml --profile reindex run --rm reindex
docker compose -f docker-compose.cloud.yml up -d app
Notes
- Pinecone Local is development-only.
- Pinecone Local is in-memory and not production-safe.
- Pinecone Local records do not persist after shutdown.
- The Docker image clones
git@github.com:J3ys/context-mcp.gitand builds directly from the fork.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。