AI Knowledge Library — structured, cited knowledge units for AI agents. Pre-verified answers that save tokens, reduce hallucinations, and cite every source.
1,800+ knowledge units across 18 domains (consumer electronics, software, business strategy, ERP integration, compliance, energy, finance, and more). Each unit answers one canonical question with:
- Confidence scores (0.0-1.0) per published methodology
- Inline source citations from 5-8 authoritative sources
- Freshness tracking with verified dates and temporal validity
- Quality status — verified, needs_review, or unreliable
- Knowledge graph — related units with typed edges
One API call replaces 5 web searches and 8,000 tokens of parsing.
npx knowledgelib-mcpOr add to claude_desktop_config.json:
{
"mcpServers": {
"knowledgelib": {
"command": "npx",
"args": ["knowledgelib-mcp"]
}
}
}POST https://knowledgelib.io/mcp
Streamable HTTP transport, JSON-RPC 2.0, MCP spec 2025-03-26.
# Search
curl https://knowledgelib.io/api/v1/query?q=best+wireless+earbuds+under+150
# Batch search (up to 10 queries)
curl -X POST https://knowledgelib.io/api/v1/batch \
-H "Content-Type: application/json" \
-d '{"queries":[{"q":"earbuds"},{"q":"headphones"}]}'
# Get full unit
curl https://knowledgelib.io/api/v1/units/consumer-electronics/audio/wireless-earbuds-under-150/2026.md
# Health check
curl https://knowledgelib.io/api/v1/healthpip install langchain-knowledgelibfrom langchain_knowledgelib import KnowledgelibRetriever
retriever = KnowledgelibRetriever()
docs = retriever.invoke("best wireless earbuds")npm install n8n-nodes-knowledgelib| Tool | Description | Read-only |
|---|---|---|
query_knowledge |
Search across all knowledge units with filters | Yes |
batch_query |
Search multiple topics in one call (max 10) | Yes |
get_unit |
Retrieve full markdown content by ID | Yes |
list_domains |
List all domains with unit counts | Yes |
suggest_question |
Submit a topic request for new unit creation | No |
report_issue |
Flag incorrect, outdated, or broken content | No |
All read-only tools are marked with readOnlyHint: true and idempotentHint: true per MCP spec 2025-03-26, enabling parallel execution by agents.
- Structured error codes with retryable flag and retry_after_ms
- ETag / If-None-Match caching (304 Not Modified)
- Correlation IDs (X-Request-Id header on all responses)
- Quality status (verified / needs_review / unreliable) on all results
- Related units for knowledge graph traversal
- Content previews (150-char summaries without fetching full unit)
- Token budgeting (total_tokens across results)
- Rate limiting on write endpoints (10 suggestions/hr, 20 feedback/hr)
- Zod validation with per-field error messages
| Type | Count | Description |
|---|---|---|
| product_comparison | 418 | Best-of roundups with decision logic and buy links |
| concept | 336 | Definitions of terms agents often get wrong |
| software_reference | 239 | Code examples, anti-patterns, decision trees |
| execution_recipe | 202 | Step-by-step implementation plans |
| erp_integration | 166 | API capabilities, rate limits, data mapping |
| agent_prompt | 55 | System prompts for pipeline sub-agents |
| assessment | 54 | Structured scoring frameworks |
| decision_framework | 35 | Decision trees with trade-offs |
| benchmark | 28 | Industry benchmarks by segment |
| rule | 28 | Actionable directives with evidence |
- /llms.txt — Plain-text guide for LLMs
- /llms-full.txt — Complete index of all questions
- /.well-known/ai-knowledge.json — Machine-readable manifest
- /catalog.json — Full catalog with metadata
- /for-agents — Integration guide
- Website: https://knowledgelib.io
- npm: https://www.npmjs.com/package/knowledgelib-mcp
- PyPI: https://pypi.org/project/langchain-knowledgelib/
- HTTP MCP: https://knowledgelib.io/mcp
- OpenAPI: https://knowledgelib.io/api/v1/openapi.json
- GPT Actions: https://knowledgelib.io/.well-known/openapi-gpt.json
CC BY-SA 4.0
