API Reference

Cortex exposes its functionality through two primary interfaces: the MCP protocol for AI agent integration and the CLI for direct interaction.

MCP Tools

16+ MCP tools for conversation, knowledge, context, and entity operations. Works with any MCP-compatible client.

CLI Reference

Complete CLI documentation for all commands, flags, and usage examples.

MCP Protocol

Cortex implements the Model Context Protocol, making memory primitives available as discoverable tools.

Transports

Transport	Use Case
`stdio`	Claude Desktop, local tools, subprocess integration
`sse`	Web clients, remote connections, HTTP-based tooling

Tool Categories

Category	Tools	Description
Conversation	4	Thread-based dialogue with semantic search
Knowledge	4	Document ingestion and hybrid search
Context	5	Key-value storage with versioning
Entity	6	Knowledge graph operations

See MCP Tools Reference for complete documentation.

Configuration

Cortex behavior is controlled via ~/.cortex/config.yaml:

storage:
  backend: sqlite
  data_dir: ~/.cortex/data

embedding:
  provider: openai          # openai, anthropic, voyageai, gemini, ollama
  model: text-embedding-3-small
  dimensions: 1536
  batch_size: 100
  cache_size: 1000

summarization:
  provider: anthropic
  model: claude-sonnet-4-6
  max_tokens: 1024

server:
  metrics_enabled: true
  metrics_port: 9811

Required Environment Variables:

# Set API key for your embedding provider
export OPENAI_API_KEY="sk-..."        # For openai provider
export ANTHROPIC_API_KEY="sk-ant-..." # For anthropic provider

Observability

Metrics Endpoint

When metrics_enabled: true, Prometheus metrics are available at :9811/metrics.

Key metrics:

cortex_operations_total - Operations by primitive/action/status
cortex_operation_duration_seconds - Latency histogram
cortex_search_latency_seconds - Search-specific timing
cortex_embedding_requests_total - Embedding API calls

Health Check

curl http://localhost:9811/health

Returns {"status": "ok"} when healthy.