ocp-server 0.3.0

Open Context Protocol — reference server implementation

ocp-server

Reference server implementation for the Open Context Protocol (OCP) — a vendor-neutral protocol for sharing retrievable context, persistent state, and invalidation events across AI agents, models, and frameworks.

Every OCP server is an MCP server. Drop it into any MCP-compatible IDE or agent runtime with a single config line.

What it does

• Semantic retrieval — index any codebase or document set, run k-NN search over embedded chunks, assemble token-budget-aware context bundles ready for LLM prompts
• Scoped state — typed key-value store with agent, session, and global scopes and optimistic concurrency control
• Session coordination — agent handoff, named checkpoints, restore from checkpoint
• Invalidation & events — file watcher marks stale chunks automatically; real-time event subscriptions with replay

Quick start

# Zero-install — run directly from PyPI
uvx ocp-server

# Or install permanently
pip install ocp-server
ocp-server          # stdio transport (for IDE / agent integrations)
ocp-server-http     # HTTP/SSE transport (for multi-client / remote scenarios)

IDE integrations

Add to .mcp.json in your project root (Claude Code, VS Code, JetBrains):

{
  "mcpServers": {
    "ocp": {
      "command": "uvx",
      "args": ["ocp-server"],
      "env": { "OCP_DB_PATH": "${workspaceFolder}/.ocp.db" }
    }
  }
}

Works the same way for Claude Desktop, Cursor, and any other MCP-compatible runtime. See the integrations guide for copy-paste configs.

Storage backends

Backend	How to enable
SQLite (default)	Zero config — just run ocp-server
PostgreSQL + pgvector	Set OCP_DATABASE_URL=postgresql://...

Embedding backends

Backend	OCP_EMBEDDER	Extra deps	Quality
Hash n-gram (default)	hash	none	keyword-level
FastEmbed (BAAI/bge)	fastembed	pip install fastembed	semantic
OpenAI	openai	pip install openai	semantic

Docker

docker pull ghcr.io/rajesh1213/ocp:latest

# Run with SQLite
docker run -p 8080:8080 -e OCP_API_KEYS=my-key ghcr.io/rajesh1213/ocp:latest

# Run with PostgreSQL
docker run -p 8080:8080 \
  -e OCP_DATABASE_URL=postgresql://user:pass@host/db \
  -e OCP_API_KEYS=my-key \
  ghcr.io/rajesh1213/ocp:latest

Configuration

All configuration via environment variables:

Variable	Default	Description
OCP_DB_PATH	ocp.db	SQLite file path
OCP_DATABASE_URL	—	PostgreSQL DSN (enables Postgres backend)
OCP_API_KEYS	—	Comma-separated Bearer tokens (empty = open/dev mode)
OCP_API_KEY	—	Key the client presents in stdio mode
OCP_HOST	0.0.0.0	HTTP server bind address
OCP_PORT	8080	HTTP server port
OCP_EMBEDDER	hash	Embedding backend: hash, fastembed, openai
OCP_WATCH	1	Set to 0 to disable file watching

Conformance

The reference server passes all 44 OCP-0002 conformance tests at the core+coordination level.

# Run conformance suite against this server
pip install ocp-conformance
ocp-conformance

Links

• GitHub
• Protocol spec (OCP-0001)
• Usage guide
• Integrations
• ocp-client — Python async client SDK

License

Apache-2.0 — see LICENSE