fbdl-mcp
MCP server for Meta's FBDL (Facebook Developer Language). Lets AI agents generate, validate, and explore FBDL scripts used in Meta's bug bounty program (MMBRC).
README
fbdl-mcp
MCP server for Meta's FBDL (Facebook Developer Language). Lets AI agents generate, validate, and explore FBDL scripts used in Meta's bug bounty program (MMBRC).
No auth required -- FBDL is a text format. This server runs locally and never talks to Meta's infrastructure.
What it does
- validate_fbdl -- Parse and validate FBDL scripts. Catches unknown entities/actions, missing required params, invalid enum values, bad syntax.
- list_entities -- Browse all setup entity types (User, Page, Group, Album, Friendship, Business, App, Event) with their params and examples.
- list_actions -- Browse all 70+ FBDL actions with signatures, params, and examples. Filter by name or category.
- explain_fbdl -- Turn an FBDL script into a plain-English step-by-step explanation.
- generate_fbdl (prompt) -- Prompt template that gives the calling LLM the full FBDL grammar so it can generate valid scripts from natural language.
Claude Code skills (no MCP server needed)
The repo includes standalone Claude Code skills that work without the MCP server. Copy the .claude/skills/ directory into any project:
cp -r /path/to/fbdl-mcp/.claude/skills/ your-project/.claude/skills/
Available skills:
/generate-fbdl-- Generate FBDL scripts from natural language. Embeds the full grammar reference so Claude can produce valid scripts directly./validate-fbdl-- Validate FBDL scripts against the grammar. Includes a complete checklist of structure, entity, action, and reference checks.
These skills embed the entire FBDL spec as prompt context -- no server, no dependencies, no build step.
Install
git clone <repo-url> && cd fbdl-mcp
npm install
npm run build
Verify everything works:
npm run check # typecheck + lint + format + tests
Setup with Claude Code
Add to your Claude Code MCP config (~/.claude/settings.json or project .claude/settings.json):
{
"mcpServers": {
"fbdl": {
"command": "node",
"args": ["/absolute/path/to/fbdl-mcp/dist/server.js"]
}
}
}
Then in Claude Code you can say things like:
"Generate an FBDL script where two users are friends, one owns a page, the other posts a photo and gets blocked"
"Validate this FBDL script: [setup] User UserOne Page PageOne with {owner: UserOne}"
"What FBDL actions are available for groups?"
Setup with OpenAI Codex / agents
For Codex or any MCP-compatible agent, point it at the server binary:
node /absolute/path/to/fbdl-mcp/dist/server.js
The server communicates over stdio using the MCP protocol. It exposes:
- 4 tools:
validate_fbdl,list_entities,list_actions,explain_fbdl - 1 resource:
fbdl://reference(full language spec as markdown) - 1 prompt:
generate_fbdl(NL-to-FBDL prompt template)
AGENTS.md snippet
If your agent framework uses AGENTS.md for tool discovery, add:
## FBDL MCP Server
Tools for working with Meta's FBDL (Facebook Developer Language) scripts:
- Use `validate_fbdl` after generating a script to check it for errors
- Use `list_entities` and `list_actions` to discover available FBDL constructs
- Use `explain_fbdl` to understand what an existing script does
- Use the `generate_fbdl` prompt to produce scripts from natural language
The server runs on stdio. Start with: `node /path/to/fbdl-mcp/dist/server.js`
Example workflow
-
Agent receives: "Create a test scenario where a page admin blocks a group member"
-
Agent calls
list_entitiesto check what setup is needed -
Agent calls
list_actionswithcategory: "block"to find the right action -
Agent uses the
generate_fbdlprompt to produce:
[setup] User OwnerOne User MemberOne Page PageOne with {owner: OwnerOne} Group GroupOne with {owner: OwnerOne, privacy: private, members: [MemberOne]}
OwnerOne as PageOne block MemberOne
-
Agent calls
validate_fbdlto verify the script is correct -
Researcher pastes the script into Meta's MMBRC platform
FBDL quick reference
Setup block
[setup] Type Label [with {key: value, ...}] [Type Label ...]
Entity types: User, Page, Group, Album, Friendship, Business, App, Event
Action lines
Subject [as VoiceSwitcher] action_name Label [with {key: value, ...}]
Voice switcher (as) lets a user act as a page they own.
Rules
- All entities must be created in setup before use in actions
- The setup block is a single line starting with
[setup] - Each action goes on its own line
- Labels are PascalCase and must be unique
Development
npm run build # compile TypeScript
npm run test # run tests
npm run lint # eslint with strict TypeScript rules
npm run format # prettier
npm run check # all of the above
Project structure
src/
schema.ts # FBDL language spec as typed data (entities + actions)
validator.ts # FBDL script parser and validator
server.ts # MCP server (tools, resources, prompts)
__tests__/
schema.test.ts # Schema integrity tests
validator.test.ts # Validator correctness tests
server.test.ts # MCP tool integration tests
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。