Langchain Docs

SKILL: LangChain Documentation Expert

[!IMPORTANT] Purpose: Access LangChain's official documentation in real-time via MCP. Always fetch fresh docs to avoid outdated information.

Documentation: docs.langchain.com

When to Use

Load this skill when:

• User asks about LangChain, LangGraph, or Deep Agents concepts
• Implementing multi-agent systems or state graphs
• Debugging LangChain/LangGraph code
• Learning about middleware, backends, or skills patterns
• Comparing Phylactery architecture with LangChain standards

Critical Patterns

1. Documentation Workflow

Step-by-step process:

1. Fetch the Documentation Index

   MCP Tool: fetch_url
   URL: https://docs.langchain.com/llms.txt
   

   This provides a structured list of all available documentation with descriptions.

2. Select Relevant Documentation Based on the user's question, identify 2-4 most relevant documentation URLs from the index.

   Prioritize:

   • Specific how-to guides for implementation questions
   • Core concept pages for understanding questions
   • Tutorials for end-to-end examples
   • Reference docs for API details

3. Fetch Selected Documentation

   MCP Tool: fetch_url
   URL: <selected_doc_url>
   

   Fetch each selected documentation page.

4. Provide Accurate Guidance After reading the documentation:

   • Answer the user's question with citations
   • Include code examples from the docs
   • Link to relevant pages for deeper reading
   • Highlight differences with Phylactery's approach (if applicable)

2. MCP Server Connection

Server Details:

• URL: https://docs.langchain.com/mcp
• Type: HTTP MCP Server (Mintlify-hosted)
• Tools Available: fetch_url, search_docs

Connection Test:

# Verify MCP connection
mcp_client = get_mcp_client("langchain-docs")
result = await mcp_client.call_tool(
    "fetch_url",
    {"url": "https://docs.langchain.com/llms.txt"}
)

3. Progressive Disclosure

This skill should only load when:

• User query contains keywords: "langchain", "langgraph", "deep agents", "middleware", "backends"
• User explicitly requests LangChain documentation
• Comparing architectures or patterns

Do NOT load for:

• General Python questions
• Angular/Frontend questions
• Questions about other frameworks

Code Examples

Example: Fetching Deep Agents Middleware Docs

# 1. Get index
index = await fetch_url("https://docs.langchain.com/llms.txt")

# 2. Search for "middleware"
relevant_docs = [
    "https://docs.langchain.com/oss/deepagents/middleware",
    "https://docs.langchain.com/oss/deepagents/harness"
]

# 3. Fetch each doc
for doc_url in relevant_docs:
    content = await fetch_url(doc_url)
    # Parse and extract relevant sections

Example: Comparing with Phylactery

When user asks: "How does LangChain handle filesystem tools?"

Response Pattern:

According to [LangChain Deep Agents Docs](https://docs.langchain.com/oss/deepagents/harness):

**LangChain Approach:**
- 6 filesystem tools: ls, read_file, write_file, edit_file, glob, grep
- Pluggable backends: StateBackend, StoreBackend, FilesystemBackend, CompositeBackend
- Automatic tool result eviction (>20k tokens)

**Phylactery Current State:**
- Uses MCP filesystem server (external)
- No native filesystem middleware yet
- Planned for Phase 2 (see ROADMAP.md)

**Recommendation:** Implement FilesystemMiddleware following LangChain's pattern.

Anti-Patterns

Don't: Use Outdated Information

❌ Never answer LangChain questions from training data alone. ✅ Always fetch fresh documentation via MCP.

Why: LangChain evolves rapidly. Deep Agents was released in late 2024.

Don't: Fetch Entire Documentation

❌ Never fetch all docs at once (hundreds of pages). ✅ Always use the index to select 2-4 most relevant pages.

Why: Context window limits and performance.

Don't: Ignore Phylactery Context

❌ Never recommend LangChain patterns without considering Phylactery's architecture. ✅ Always explain how to adapt LangChain concepts to Phylactery.

Why: We're building on LangChain ideas, not copying blindly.

Quick Reference

Common Queries

MCP Tools

Key Concepts to Know

• Deep Agents: LangChain's agent harness with built-in tools
• Middleware: Composable capabilities (TodoList, Filesystem, SubAgent)
• Backends: Storage abstraction (State, Store, Filesystem, Composite)
• Skills: Reusable agent capabilities (Agent Skills standard)
• Progressive Disclosure: Load context only when needed

Last Updated: 2026-01-26
Maintained By: SkullRender AI (Phylactery Team)