What is the Contox MCP Server?
The Contox MCP Server (@contox/mcp-server) is a Model Context Protocol server that gives Claude and other MCP-compatible AI assistants persistent project memory. It exposes 23 tools that let the AI load, save, search, and manage structured project knowledge across sessions.
The server binary is contox-mcp and communicates over stdio using the MCP protocol.
- Package:
@contox/mcp-server - Binary:
contox-mcp - Version: 3.5.0
- Protocol: Model Context Protocol (MCP) over stdio
All 23 Tools
The MCP server provides 23 tools organized into five categories:
Memory lifecycle
| Tool | Description |
|---|---|
contox_get_memory | Load the complete project memory brain document at session start |
contox_save_session | Save session work into persistent memory (user-initiated only) |
contox_context_pack | Build a focused, token-budgeted context pack for a specific task |
CRUD operations
| Tool | Description |
|---|---|
contox_list_contexts | List all contexts for the current project |
contox_get_context | Get a specific context by ID with full content |
contox_create_context | Create a new context manually |
contox_update_context | Update an existing context (name, description, or content) |
contox_delete_context | Delete a context by ID (irreversible) |
Brain management
| Tool | Description |
|---|---|
contox_search | Semantic search across all memory via V2 embeddings |
contox_populate | Batch create/update contexts by schemaKey |
contox_add_entry | Add an entry to a journal-like context (event-sourced) |
contox_get_tree | Get the hierarchical brain tree |
contox_approve | Approve or deprecate a context (state machine) |
Knowledge graph
| Tool | Description |
|---|---|
contox_get_links | Get incoming/outgoing links for a context |
contox_add_link | Create a directed link between two contexts |
Operations
| Tool | Description |
|---|---|
contox_compact | Compact journal entries into a summary |
contox_oncall_view | View on-call operational summary and health stats |
contox_explain_schemakey | Deep-dive into any brain schemaKey |
contox_scan | Scan a codebase and push hierarchical sub-contexts |
contox_git_digest | Read git commits since last save for enrichment |
contox_hygiene | Memory hygiene agent — analyze and clean up memory items |
contox_ask | Ask a natural-language question about your project memory |
contox_auto_resolve | Automatically resolve memory items when code fixes are committed |
How the AI uses these tools
A typical session follows this pattern:
- Session start — The AI calls
contox_get_memoryto load the full project brain - During work — The AI uses
contox_search,contox_context_pack, and other tools as needed - Session end — When the user explicitly asks, the AI calls
contox_save_sessionto persist the session's work
The MCP server handles all communication with the Contox API. The AI never needs to make HTTP requests directly.
Next steps
- Installation — Install and verify the MCP server
- Configuration — Configure for Claude Desktop,
.mcp.json, and environment variables