@astrivya/mcp-server

Model Context Protocol server for the Astrivya Knowledge Graph. Provides 18 tools that any MCP-compatible client (Claude Desktop, Cursor, Cline, VS Code with Continue, etc.) can use to read, write, and search your local knowledge graph.

npm install -g @astrivya/mcp-server

# Run with stdio transport (default for Claude Desktop)
astrivya-mcp

# Or as HTTP/SSE on port 3001
astrivya-mcp --http --port 3001

Setup

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "astrivya": {
      "command": "npx",
      "args": ["-y", "@astrivya/mcp-server"]
    }
  }
}

VS Code with Continue

Add to ~/.continue/config.json:

{
  "experimental": {
    "mcpServers": {
      "astrivya": {
        "command": "npx",
        "args": ["-y", "@astrivya/mcp-server"]
      }
    }
  }
}

Tools

| Tool | Input | Description | |------|-------|-------------| | search_memories | query, limit? | Search the knowledge graph with semantic ranking | | get_team_context | team_id? | Workspace context (nodes, edges, chunks stats) | | get_person_context | — | Current workspace status | | get_mcp_status | — | MCP server health: uptime, sessions, tool call counters, recent events | | log_decision | title, content, file_context? | Record a design decision | | log_memory | content, type?, visibility? | Store a custom fact or insight | | search_connectors | query, limit? | Search indexed content chunks | | get_daily_briefing | limit? | Recent workspace activity | | find_related_knowledge | term, limit? | Semantic search for related concepts | | get_expertise_profile | team_id? | Workspace knowledge composition | | trace_decision | decision_id | Trace a decision's graph neighborhood | | get_team_members | team_id? | (Requires Astrivya Cloud API) | | get_team_analytics | team_id? | (Requires Astrivya Cloud API) | | list_notifications | unread?, type?, limit? | (Requires Astrivya Cloud API) | | mark_notification_read | notification_id? | (Requires Astrivya Cloud API) | | check_credits | — | Live credit balance, lifetime usage, recent transactions (cloud; fails gracefully offline) | | get_context_digest | limit? | Compact pre-digested context for session start (prose) | | get_workspace_updates | since?, limit? | Knowledge-graph changes since a timestamp (delta) |

Cloud-requiring tools gracefully fall back to local AKG stats when the cloud API is unavailable.

Resources

| URI | Description | |-----|-------------| | astrivya://team/active/knowledge | Workspace metadata | | astrivya://user/active/memories | Personal memory summary | | astrivya://briefing/today | Recent activity | | astrivya://org/knowledge-graph | All nodes in the graph |

Environment

| Variable | Description | |----------|-------------| | ASTRIVYA_WORKSPACE_PATH | Workspace root (default: cwd) | | ASTRIVYA_TOKEN | Cloud API token (optional, for sync) | | ASTRIVYA_SYNC_KEY | Sync API key (optional) | | ASTRIVYA_CLOUD_URL | Cloud API URL (optional) |

Telemetry

The MCP server sends anonymous, opt-out usage stats (server starts/stops, session start/end, tool names, client name, and a random install id) to PostHog so we can improve the product. No code, paths, arguments, query content, tokens, or any other sensitive data is ever collected. Tool calls are sampled at most once per session per tool.

• You'll see a one-time notice on stderr at startup.
• Disable with ASTRIVYA_TELEMETRY=off or NO_TELEMETRY=1.
• Telemetry is automatically disabled in CI environments.

License

Apache 2.0