AT Protocol MCP Server

Appearance

AT Protocol MCP ServerComprehensive LLM Integration

Enable LLMs to interact directly with the AT Protocol ecosystem through a powerful Model Context Protocol server

Get Started
View on GitHub
AT Protocol MCP Server Logo

Complete AT Protocol Integration

51 tools built on the official @atproto/api, covering posts, replies, likes, reposts, follows, profiles, timelines, search, direct messages, bookmarks, starter packs, and moderation.

MCP Server Compliance

Built with @modelcontextprotocol/sdk following the official MCP specification, ensuring compatibility with all MCP-compatible LLM clients.

Type-Safe & Secure

Written in TypeScript with strict type checking, comprehensive input validation, and per-tool rate limiting.

Batch Operations

Perform multiple operations in a single call - follow, like, or repost up to 25 items at once to reduce round-trips and improve efficiency.

Analytics & Insights

Summarize engagement velocity and map network connections using heuristics over your own posts and graph data - no ML, just transparent metrics.

Content Discovery

Find similar users from shared follows/followers and sample trending topics from your home timeline.

Resources & Prompts

Three MCP resources (timeline, profile, notifications), two parameterized resource templates (any actor's public profile and feed), and two content-composition prompts.

App Password Authentication

Authenticate with a Bluesky app password to unlock write operations, or run unauthenticated for public read-only tools.

Extensible Architecture

Modular design allows easy customization and extension with custom tools, resources, and prompts for specific use cases.

Comprehensive Documentation

Detailed documentation with examples, API reference, troubleshooting guides, and best practices for integration.

What's New in v0.6.0 ​

The latest release grows the roster from 43 to 51 tools and broadens the MCP surface:

New Tool Areas

Bluesky direct messages (list_conversations, get_conversation_messages, send_direct_message), private bookmarks (add_bookmark, remove_bookmark, get_bookmarks), and starter packs (search_starter_packs, get_starter_pack).

Reply & Quote Controls

create_post and create_thread can gate who may reply (a threadgate record via replyControls) and whether a post may be quoted (a postgate record via quoteControls).

Richer MCP Surface

Parameterized resource templates (atproto://profile/{actor}, atproto://feed/{actor}) serve any actor's public data — no authentication required — and the server advertises the completions capability for prompt arguments and template variables.

Streamable HTTP Transport

--transport http serves MCP over Streamable HTTP at /mcp (loopback by default). stdio remains the default transport for MCP clients.

Previously in v0.4.0 ​

v0.4.0 reshaped the tool API for clarity and reliability, consolidating the 62-tool roster down to 43 tools that are all functional, single-purpose, and fully documented:

Consolidated Toolset

Redundant tools were merged behind parameters — for example get_followers + get_follows became get_user_connections (direction: 'followers' | 'follows'), and the batch tools became one batch_action (action: 'follow' | 'like' | 'repost'). Tools that never worked (OAuth flows, real-time streaming, generate_alt_text) were removed.

Machine-Readable Schemas

Every tool now advertises a JSON Schema outputSchema in tools/list, and every tool parameter carries a description. A test enforces 100% coverage so this cannot regress.

New Tools in 0.4.0

search_actors finds accounts by handle or display name, and get_author_feed lists a specific user's posts.

Removed Experimental Tools

The non-functional OAuth and real-time streaming tools were removed in 0.4.0; they are no longer registered. See Experimental & Roadmap for current status.

Quick Start ​

Get up and running with the AT Protocol MCP Server in minutes:

bash
# Install globally
npm install -g atproto-mcp

# Start the server
atproto-mcp

# Or use with npx
npx atproto-mcp

Key Features ​

Social Operations ​

  • Create posts with rich text, images, and embeds
  • Reply to posts with proper threading
  • Like, repost, and manage reactions
  • Follow and unfollow users
  • Access timelines and feeds

Data Retrieval ​

  • Search posts and content across the network
  • Retrieve user profiles and information
  • Access follower and following lists
  • Get notifications
  • Read threads, custom feeds, and timelines
  • Send and read Bluesky direct messages
  • Save and list private bookmarks
  • Search and inspect starter packs

Authentication ​

  • App password authentication for write operations
  • Unauthenticated mode for public read-only tools
  • In-memory session handling

OAuth login is not yet implemented and is not exposed as a tool. Real-time firehose streaming is not planned as tools — see Experimental & Roadmap.

Developer Experience ​

  • TypeScript with full type safety
  • Comprehensive error handling
  • Detailed logging via the --log-level flag
  • Extensive test coverage

Architecture ​

The AT Protocol MCP Server bridges the gap between LLMs and the AT Protocol ecosystem:

Use Cases ​

This MCP server enables LLMs to provide powerful AT Protocol capabilities through natural language interaction:

  • LLM-Powered Social Automation: Enable AI assistants to manage posting, engagement, and content on behalf of users through conversational commands
  • AI-Driven Content Analysis: Allow LLMs to analyze social media trends and user behavior through natural language queries and provide insights
  • Intelligent Community Management: Empower LLMs to help users manage communities, moderate content, and engage with followers
  • Conversational Data Integration: Let LLMs integrate AT Protocol data with other systems through natural language workflows
  • AI-Assisted Research & Analytics: Enable LLMs to conduct social media research and analytics by querying and processing AT Protocol data
  • Smart Bot Development: Build intelligent social media bots where LLMs handle natural language understanding and this server handles AT Protocol actions

How it works: Users interact with their LLM client (like Claude Desktop) in natural language, and the LLM uses this MCP server to access AT Protocol functionality. For example, a user might say "Search for trending posts about AI and summarize them," and the LLM would use the search_posts tool to fulfill that request.

Community ​

Join our growing community of developers building with the AT Protocol:

License ​

Released under the MIT License.

Made with ❤️ by Cameron Rye