Lessons Learned MCP Server

Lessons Learned MCP Server

Central MCP server for storing and querying lessons learned and project-specific implementation details, compatible with MCP-capable AI clients like Cursor, Claude Code, and Google Antigravity.

Category
访问服务器

README

Lessons Learned MCP Server

Central MCP server for storing and querying lessons learned and project-specific implementation details. A Laravel application that exposes two MCP endpoints for any MCP-capable AI client — Cursor, Claude Code, Google Antigravity, and others.

Explainer Video

MCP Server explainer video

Watch the explainer on YouTube.

Use with ashwinmram/mcp-pusher

This server works with the ashwinmram/mcp-pusher package to push lessons and project details from your Laravel projects via HTTP API.

  • Install: composer require ashwinmram/mcp-pusher:^3.0
  • Links: GitHub | Packagist
  • Best practice: Copy the knowledge capture prompt into your agent; agent gathers git context, synthesizes lessons, runs mcp:append → drafts. mcp:push once at session end. Optional Cursor preCompact hook surfaces the same prompt automatically.

The mcp-pusher package pushes session drafts from docs/.mcp-session/*.jsonl to /api/lessons (generic) and /api/project-details (project-specific) in one mcp:push. See Pushing knowledge (mcp-pusher 3.0) and packages/laravel-mcp-pusher/README.md.

Key Features

  • Lessons Learned MCP (/mcp/lessons) — Search, browse, and retrieve lessons with relevance scoring
  • Project Details MCP (/mcp/project-details?project=…) — Project-specific implementation details (paths, env, conventions)
  • Tools: SearchLessons, GetLessonByCategory, GetLessonTags, FindRelatedLessons, MarkLessonHelpful, SuggestSearchQueries, GetTopLessons, GetCategoryStatistics
  • Resources: lessons://overview, lessons://search-guide
  • Prompts: LessonsLearnedOverview, LessonsByCategory

Dashboard

Authenticated users can view live stats at /dashboard. Each card shows a current value, a baseline, and a variance badge.

Two comparison patterns are used, both anchored on 30 days:

Metric Current value Baseline (gray text) Badge change
Total Lessons, Total Project Details, Source Projects, per-source-project cards Cumulative count now Same count at 30 days ago Percent change
Retrievals, Details Added Count in last 30 days Count in prior 30 days (days 30–60 ago) Percent change
Helpfulness Rate Rate in last 30 days Rate in prior 30 days Percentage-point change (e.g. 60% vs 50% = +10.0 points)

See in-app Documentation for the same details.

Prerequisites

  • PHP 8.2+, Laravel 12
  • MySQL or SQLite
  • Laravel Herd (recommended) or equivalent local dev setup
  • An MCP-capable AI client (Cursor, Claude Code, Google Antigravity, or other)

Initial Setup

1. Clone and Install

composer install
cp .env.example .env
php artisan key:generate
php artisan migrate

2. Configure Environment Variables

Add the following to your .env file:

# MCP server (HTTP base URL for MCP clients and API pushes)
APP_URL=https://mcp-server.test
MCP_SERVER_NAME="Lessons Learned MCP Server"

# MCP client (for pushing from this project via mcp-pusher)
MCP_SERVER_URL=https://mcp-server.test
MCP_API_TOKEN=  # Generated in step 3

3. Generate API Token

Generate a Sanctum API token for MCP authentication:

php artisan mcp:generate-token --name="mcp-client-token"

This command creates a user if none exists, generates a Sanctum token, and displays it. Copy the token (shown only once) and add it to .env:

MCP_API_TOKEN=your-token-here

Use the same token in your MCP client Authorization: Bearer … header.

4. Verify Setup

Ensure your site is accessible. With Laravel Herd:

curl https://mcp-server.test

If needed, link the project to Herd:

herd link mcp-server

Test the MCP endpoint:

curl -H "Authorization: Bearer YOUR_TOKEN_HERE" \
     https://mcp-server.test/mcp/lessons

Configuring your AI client

All clients need:

  • Lessons URL: https://mcp-server.test/mcp/lessons
  • Project details URL (optional): https://mcp-server.test/mcp/project-details?project=<source>project must match php artisan mcp:push --source=…
  • Header: Authorization: Bearer YOUR_SANCTUM_TOKEN

Example config files (copy and edit): cursor, claude, antigravity. Root mcp.json.example is a minimal Cursor-oriented sample.

Cursor

Settings UI

  1. Open Cursor Settings: ⇧+⌘+J (Mac) or Ctrl+Shift+J (Windows/Linux)
  2. FeaturesMCPAdd MCP Server
  3. Configure:
    • Name: lessons-learned-local (or any name)
    • Transport: SSE
    • URL: https://mcp-server.test/mcp/lessons
    • Headers: {"Authorization": "Bearer YOUR_SANCTUM_TOKEN_HERE"}
  4. Save and restart Cursor

Project mcp.json

{
    "mcpServers": {
        "lessons-learned-local": {
            "transport": "sse",
            "url": "https://mcp-server.test/mcp/lessons",
            "headers": {
                "Authorization": "Bearer YOUR_SANCTUM_TOKEN_HERE"
            }
        }
    }
}
  • Global: ~/.cursor/mcp.json or Settings UI — all projects
  • Project: mcp.json in project root — this project only (gitignore tokens)

Verify: Settings → MCP → server Connected. Ask: "What lessons do we have about testing Laravel packages?"

Claude Code

Prefer HTTP transport (Laravel Mcp::web uses HTTP). Use SSE only if HTTP fails.

CLI

claude mcp add --transport http lessons-learned \
  https://mcp-server.test/mcp/lessons \
  --header "Authorization: Bearer YOUR_TOKEN"

claude mcp add --transport http project-details-my-app \
  "https://mcp-server.test/mcp/project-details?project=my-app" \
  --header "Authorization: Bearer YOUR_TOKEN"

Project .mcp.json

{
    "mcpServers": {
        "lessons-learned": {
            "type": "http",
            "url": "https://mcp-server.test/mcp/lessons",
            "headers": {
                "Authorization": "Bearer YOUR_TOKEN"
            }
        }
    }
}

Scopes: local (default, ~/.claude.json), project (.mcp.json in repo), or user. See Claude Code MCP docs.

Verify: claude mcp list — servers show connected; start a session and query lessons.

Google Antigravity

Shared config: ~/.gemini/config/mcp_config.json (Antigravity IDE and CLI share this path per Google codelabs).

{
    "mcpServers": {
        "lessons-learned": {
            "serverUrl": "https://mcp-server.test/mcp/lessons",
            "headers": {
                "Authorization": "Bearer YOUR_TOKEN"
            }
        },
        "project-details-my-app": {
            "serverUrl": "https://mcp-server.test/mcp/project-details?project=my-app",
            "headers": {
                "Authorization": "Bearer YOUR_TOKEN"
            }
        }
    }
}

UI: Agent panel → ...MCP Servers → refresh; add or verify servers.

Verify: MCP server shows connected; ask the agent about available lessons.

Agent startup instructions

Install agent instructions for your AI client:

Client Command
Cursor php artisan mcp:install-cursor-rules
Claude Code php artisan mcp:install-claude-instructions
Google Antigravity php artisan mcp:install-antigravity-skills
All php artisan mcp:install-agent-instructions

See agent-instructions README for flags and manual paths.

# Example: full setup for Cursor + Claude + Antigravity (workspace skills)
php artisan mcp:install-agent-instructions --with-hooks --with-claude-md --with-cursorrules

This monorepo uses a short .cursorrules index; consumer projects should use the install commands above or copy stubs from packages/laravel-mcp-pusher/stubs/.

How sessions use the MCP server

  1. Your AI client connects with the configured URL and Bearer token
  2. The server exposes lessons://overview and MCP tools/prompts
  3. With startup instructions configured, the agent reads the overview and queries relevant lessons
  4. Lessons inform coding decisions during the session

Example prompts (any client)

  • "What lessons do we have about Laravel validation?"
  • "Show me lessons tagged with 'php' and 'best-practices'"
  • "Give me an overview of available lessons"
  • "What do we know about testing Laravel packages?"

Available MCP Components

Tools

  • SearchLessons — Search by keyword, category, or tags (MySQL FULLTEXT + relevance scoring)
  • GetLessonByCategory — Get all lessons in a category
  • GetLessonTags — List available tags
  • FindRelatedLessons — Find related lessons by topic relationships
  • MarkLessonHelpful — Provide feedback for relevance scoring
  • SuggestSearchQueries — Expand searches with related queries
  • GetTopLessons — Get highest relevance lessons (optionally by category)
  • GetCategoryStatistics — Category stats (avg relevance, top lessons, usage)

Resources

  • lessons://overview — Overview of lessons, categories, tags, and recent lessons
  • lessons://search-guide — Search strategies, query examples, relevance scoring

Prompts

  • LessonsLearnedOverview — Overview of available lessons
  • LessonsByCategory — Summary of lessons in a specific category

Pushing knowledge (mcp-pusher 3.0)

Consumer projects use ashwinmram/mcp-pusher. See What's new in 3.0.

Before (1.x / 2.x) After (3.0)
Edit markdown/JSON in docs/, then push mcp:append → draft JSONL during session
mcp:push-lessons + mcp:push-project-details mcp:push once (both APIs)

Workflow (all IDEs)

  1. Copy the knowledge capture prompt block from the package README into your agent (or use the optional Cursor preCompact hook).
  2. Agent gathers git context, synthesizes lessons, runs mcp:appenddocs/.mcp-session/lessons-draft.jsonl and/or project-details-draft.jsonl.
  3. End of session: review drafts (see package README), then php artisan mcp:push --source=<project>.

Draft files are cleared after a successful push unless --no-truncate. Use mcp:extract-session only if drafts are thin after compaction and you have committed work (git-only; see mcp-pusher README).

File layout (3.0)

your-project/
└── docs/
    └── .mcp-session/
        ├── lessons-draft.jsonl         ← generic lessons
        └── project-details-draft.jsonl ← project-specific details

This server (monorepo)

php artisan mcp:push --source=mcp-server

Other Laravel projects

composer require ashwinmram/mcp-pusher:^3.0

Configure MCP_SERVER_URL, MCP_API_TOKEN, gitignore docs/.mcp-session/. Full guide: package README.

Optional: Cursor hooks (this monorepo)

mkdir -p .cursor/hooks
cp packages/laravel-mcp-pusher/stubs/cursor-hooks/hooks.json.example .cursor/hooks.json
cp packages/laravel-mcp-pusher/stubs/cursor-hooks/pre-compact-checkpoint.sh .cursor/hooks/
cp packages/laravel-mcp-pusher/stubs/knowledge-capture-prompt.txt .cursor/hooks/
chmod +x .cursor/hooks/pre-compact-checkpoint.sh

Project Details MCP Server

URL: https://mcp-server.test/mcp/project-details?project=<source_project>

Use the same Bearer token. Add one MCP server entry per project (?project=… must match --source on push).

Managing Tokens

php artisan mcp:list-tokens
php artisan mcp:list-tokens --revoke=<TOKEN_ID>
php artisan mcp:list-tokens --revoke-all
php artisan mcp:generate-token --name="mcp-client-token" --force

Production Setup

  1. Update .env with production APP_URL and MCP_SERVER_URL
  2. php artisan mcp:generate-token --name="mcp-production" --email="admin@yourdomain.com"
  3. Update each MCP client with production URLs and token
  4. Set SANCTUM_STATEFUL_DOMAINS for your production domain

Troubleshooting

Token not working

  • php artisan mcp:list-tokens
  • php artisan mcp:generate-token --force

Client cannot connect

  • Confirm URL is reachable: curl -H "Authorization: Bearer TOKEN" https://mcp-server.test/mcp/lessons
  • Check Authorization: Bearer YOUR_TOKEN (include Bearer)
  • Cursor: restart app; verify Settings → MCP → Connected; SSE transport in config
  • Claude Code: claude mcp list; try --transport http then --transport sse if needed
  • Antigravity: refresh MCP Servers in Agent panel; verify ~/.gemini/config/mcp_config.json uses serverUrl and headers
  • Herd: site linked and running (herd link mcp-server)

Lessons not loading in agent

  • Restart the AI client (MCP often connects at startup)
  • Ask: "Use the LessonsLearnedOverview prompt"
  • tail -f storage/logs/laravel.log

Authentication errors (401)

  • Regenerate token; verify Bearer header in client config

SSL / certificate issues

Herd .test domains use local certificates. If the client rejects HTTPS:

  • Trust the certificate in the client, or use mkcert

Database / CORS

  • php artisan migrate:status / php artisan migrate
  • Ensure domain is in config/sanctum.php stateful domains

Security Considerations

  1. Token storage — Never commit tokens; use .env
  2. Token rotation — Rotate if compromised
  3. HTTPS — Required in production
  4. Rate limiting — Consider for MCP endpoints

Project Structure

  • routes/ai.php — MCP route registration
  • app/Mcp/ — LessonsServer, ProjectDetailsServer
  • docs/.mcp-session/ — session drafts for mcp-pusher
  • packages/laravel-mcp-pusherashwinmram/mcp-pusher (local copy)

Additional Resources

License

MIT

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选