pocket-mcp
Turns PocketBase into a persistent knowledge base, enabling Claude Code to capture and retrieve project knowledge, tasks, decisions, and workflows via semantic search.
README
Pocket MCP
A Model Context Protocol (MCP) server that turns PocketBase into a "Second Brain" for Claude Code. Capture knowledge, decisions, bugs, patterns, and more - then recall them with semantic search.
Features
- Session Management - Track work sessions with goals and outcomes
- Task Tracking - Manage project tasks with priorities, features, and progress stats
- Knowledge Capture - Save observations, decisions, bugs, code snippets
- Workflow Automation - Save and reuse repeatable procedures
- Semantic Search - AI-powered search using OpenAI embeddings
- Project Context - Quick access to project tech stack, decisions, patterns
- Knowledge Graph - Link related entities together
- Retrospectives - Generate insights from past sessions
New to the tools? Check out the Complete Usage Guide for detailed workflows and best practices.
Quick Start
Option 1: Claude Code Plugin (Recommended)
# 1. Add marketplace
/plugin marketplace add JaoChai/pocket-mcp
# 2. Install plugin
/plugin install pocket-mcp@jaochai-plugins
# 3. Configure (will prompt for credentials)
Option 2: Manual MCP Setup
Add to your project's .mcp.json:
{
"mcpServers": {
"pocket-mcp": {
"command": "npx",
"args": ["-y", "github:JaoChai/pocket-mcp"],
"env": {
"POCKETBASE_URL": "https://your-pocketbase-instance.com",
"POCKETBASE_EMAIL": "admin@example.com",
"POCKETBASE_PASSWORD": "your-password",
"OPENAI_API_KEY": "sk-..."
}
}
}
}
Then enable in .claude/settings.local.json:
{
"enableAllProjectMcpServers": true
}
Setup PocketBase
Deploy PocketBase (self-host or use PocketHost, Elestio, etc.)
Then run the setup script to create collections:
git clone https://github.com/JaoChai/pocket-mcp.git
cd pocket-mcp
npm install
# Set environment variables
export POCKETBASE_URL="https://your-instance.com"
export POCKETBASE_EMAIL="admin@example.com"
export POCKETBASE_PASSWORD="your-password"
# Run setup
npm run setup-collections
Quick Reference: When to Use Which Tool
| Event | Tools to Use |
|---|---|
| Start your day | get_pending_outcomes, get_lessons, get_project_context |
| Start new work | create_session, create_task, update_task(in_progress) |
| Need to find info | search_knowledge, semantic_search, find_workflow |
| Made a decision | capture_decision |
| Fixed a bug | capture_bug |
| Learned something new | capture_observation |
| Wrote reusable code | save_snippet |
| Finished work | update_task(done), generate_retrospective, end_session |
See Complete Usage Guide for detailed workflows.
Available Tools (28 tools)
Session Tools (3)
Manage work sessions for tracking activities.
| Tool | Description |
|---|---|
create_session |
Start a new work session with project and goal |
end_session |
End session with outcome (success/partial/failed) and summary |
get_current_session |
Get active session info |
Task Tools (5)
Track project tasks with priorities and progress.
| Tool | Description |
|---|---|
create_task |
Create a new task with title, description, feature, priority |
update_task |
Update task status, priority, or details |
get_tasks |
List tasks with filters (project, status, feature, priority) |
get_project_progress |
Get completion stats, tasks by feature, next-up items |
delete_task |
Delete or cancel a task (soft delete by default) |
Task Statuses: pending, in_progress, done, blocked, cancelled
Priorities: critical, high, medium, low
Capture Tools (4)
Save knowledge and learnings.
| Tool | Description |
|---|---|
capture_observation |
Save discoveries, patterns, insights, notes |
capture_decision |
Log decisions with context, options, and rationale |
capture_bug |
Record bugs with error message, solution, prevention |
save_snippet |
Store reusable code snippets with language and use cases |
Outcome Tracking Tools (2)
Track decision outcomes for learning.
| Tool | Description |
|---|---|
record_decision_outcome |
Record what happened after a decision was made |
get_pending_outcomes |
Find decisions older than N days that need outcome review |
Workflow Tools (4)
Save and reuse repeatable procedures.
| Tool | Description |
|---|---|
save_workflow |
Save a workflow with trigger, steps, and success criteria |
find_workflow |
Find workflow by describing what you want to do |
get_workflow |
Get full details including all steps |
record_workflow_execution |
Record that a workflow was executed |
Search Tools (2)
Find knowledge in your Second Brain.
| Tool | Description |
|---|---|
search_knowledge |
Full-text search across all collections |
semantic_search |
AI-powered similarity search using embeddings |
Context Tools (3)
Quick access to project and user context.
| Tool | Description |
|---|---|
get_project_context |
Get tech stack, recent decisions, bugs, patterns |
get_preferences |
Get user preferences (coding style, tools, workflow) |
save_preference |
Save a user preference with strength level |
Relation Tools (3)
Build a knowledge graph.
| Tool | Description |
|---|---|
link_entities |
Create relationship between two entities |
get_relations |
Get all relationships for an entity |
suggest_relations |
Auto-suggest relations based on shared tags |
Relation Types: uses, caused, fixed_by, led_to, related_to, depends_on, inspired_by
Reflection Tools (2)
Learn from past experiences.
| Tool | Description |
|---|---|
generate_retrospective |
Create retrospective with lessons and action items |
get_lessons |
Get lessons from past retrospectives |
PocketBase Collections (14)
The setup script creates these collections:
| Collection | Description |
|---|---|
projects |
Project information and tech stack |
sessions |
Work session tracking |
tasks |
Task tracking with priorities and features |
areas |
PARA method areas of responsibility |
observations |
Discoveries, patterns, insights, notes |
decisions |
Decision logs with rationale and outcomes |
bugs_and_fixes |
Error tracking and solutions |
patterns |
Design and architecture patterns |
code_snippets |
Reusable code snippets |
workflows |
Saved procedures and workflows |
retrospectives |
Session reviews and lessons |
resources |
Reference materials |
relationships |
Knowledge graph edges |
user_preferences |
User preferences |
embeddings |
Vector embeddings for semantic search |
Environment Variables
| Variable | Required | Description |
|---|---|---|
POCKETBASE_URL |
Yes | PocketBase instance URL |
POCKETBASE_EMAIL |
Yes | Admin email for authentication |
POCKETBASE_PASSWORD |
Yes | Admin password |
OPENAI_API_KEY |
Yes | OpenAI API key for embeddings |
LOG_LEVEL |
No | Log level (default: info) |
Usage Examples
Session Management
# Start a session
Claude, start a session for "implementing user authentication"
# End with summary
Claude, end the session - we completed JWT auth and refresh tokens
Task Tracking
# Create tasks
Claude, create a task "Add password reset flow" with high priority
# Check progress
Claude, show me the project progress
# Update status
Claude, mark the authentication task as done
Knowledge Capture
# Capture observation
Claude, save this: "React Query handles cache invalidation automatically in mutation's onSuccess"
# Capture decision
Claude, record that we chose JWT over sessions because of scalability needs
# Capture bug fix
Claude, save this bug: TypeError on undefined - fixed by adding optional chaining
Workflow Automation
# Save a workflow
Claude, save this deployment workflow:
1. Run tests
2. Build the project
3. Deploy to staging
4. Run smoke tests
# Find a workflow
Claude, how do I deploy to production?
# Record execution
Claude, I just ran the deployment workflow successfully
Search Knowledge
# Full-text search
Claude, search for "authentication patterns"
# Semantic search
Claude, find knowledge related to "handling API errors gracefully"
Project Context
# Get project context
Claude, what do you know about this project?
# Get preferences
Claude, what are my coding preferences?
Retrospectives
# Generate retrospective
Claude, generate a retrospective for today's session
# Get lessons
Claude, show me lessons from past sessions
Installation Options
Option 1: Plugin (Easiest)
/plugin marketplace add JaoChai/pocket-mcp
/plugin install pocket-mcp@jaochai-plugins
Option 2: npx from GitHub
{
"mcpServers": {
"pocket-mcp": {
"command": "npx",
"args": ["-y", "github:JaoChai/pocket-mcp"],
"env": { ... }
}
}
}
Option 3: From Source
git clone https://github.com/JaoChai/pocket-mcp.git
cd pocket-mcp
npm install
npm run build
Development
npm install # Install dependencies
npm run build # Build TypeScript
npm run dev # Development mode with tsx
npm run setup-collections # Setup PocketBase collections
npm test # Run tests
Architecture
pocket-mcp/
├── src/
│ ├── index.ts # Entry point (bin)
│ ├── server.ts # MCP server with 28 tools
│ ├── config.ts # Config validation (Zod)
│ ├── pocketbase/
│ │ ├── client.ts # PocketBase client
│ │ └── collections.ts # Collection schemas
│ ├── tools/
│ │ ├── session.ts # Session management
│ │ ├── tasks.ts # Task tracking
│ │ ├── capture.ts # Knowledge capture
│ │ ├── workflow.ts # Workflow automation
│ │ ├── search.ts # Full-text & semantic search
│ │ ├── context.ts # Project context & preferences
│ │ ├── relate.ts # Knowledge graph relations
│ │ └── reflect.ts # Retrospectives & lessons
│ └── utils/
│ ├── embeddings.ts # OpenAI embeddings
│ └── logger.ts # Logging (stderr for MCP)
├── scripts/
│ └── setup-collections.ts
└── dist/ # Compiled output
License
MIT
Contributing
Contributions are welcome! Please open an issue or submit a pull request.
Credits
Built with:
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。