scry

Federated search orchestrator over MCP. Query Slack, Confluence, email, and more from a single CLI — get synthesized answers with source attribution.

Quick Start

npm install -g scry
scry init
scry "what did we decide about pricing?"

How It Works

query → discover sources → parallel search → normalize → synthesize → cited answer

1. Discover: Detects MCP servers from your Claude config or PATH
2. Route: Uses a context registry (people, projects, channels) to target the right sources
3. Search: Fans out parallel queries with per-source timeouts
4. Synthesize: LLM combines results with source citations

Supported MCP Servers

| Name | Command | Install | |------|---------|---------| | Slack | slack-mcp | uv tool install git+https://github.com/aviralv/slack-mcp | | Microsoft 365 | ms365-intent-mcp | uv tool install git+https://github.com/aviralv/ms365-intent-mcp | | Confluence & Jira | confluence-jira-mcp | uv tool install git+https://github.com/aviralv/confluence-jira-mcp |

Any MCP server with search tools works — these three are bundled with optimized normalizers.

Configuration

scry init generates a scry.config.yaml:

llm:
  base_url: "https://api.anthropic.com"
  auth_token: "${ANTHROPIC_API_KEY}"
  model: "claude-haiku-4-5-20251001"

mcp_servers:
  slack:
    command: "slack-mcp"
  ms365:
    command: "ms365-intent-mcp"

search_tools:
  slack:
    - tool: "slack_search"
      params: { format: "json" }
      normalizer: "slack"
  ms365:
    - tool: "outlook_list_messages"
      params: { format: "json" }
      normalizer: "email"

registry:  # optional — enables context-aware routing
  projects:
    my-project:
      name: My Project
      routing:
        slack_channels: [team-channel]
        jira_project: PROJ

CLI Options

scry [query]              Search and synthesize
scry init                 Interactive setup wizard
scry config show          Show current configuration

Options:
  -c, --config <path>     Config file (default: scry.config.yaml)
  -t, --timeout <ms>      Per-source timeout (default: 15000)
  --no-synthesize         Show raw results without LLM synthesis

Environment Variables

| Variable | Purpose | |----------|---------| | SCRY_CONFIG | Custom config file path | | ANTHROPIC_API_KEY | LLM API key (used via ${...} in config) |

How Normalizers Work

Each MCP server returns different JSON shapes. Scry uses normalizers to convert them into a common format:

• Built-in: slack, confluence, email — optimized for known response shapes
• Generic fallback: Best-effort extraction for unknown servers (results marked low-confidence)
• Config-driven: Set normalizer: "slack" in search_tools to assign a normalizer to any tool

Requirements

• Node.js >= 20
• At least one MCP server installed and authenticated
• An Anthropic API key (for synthesis)