@sprigr/mcp-server

MCP server for AI agents with WASM-powered local full-text search and hybrid semantic search.

Full-text searches run locally via WebAssembly (~5ms). Hybrid semantic search combines keyword matching with vector embeddings for higher relevance. Writes go to the Sprigr API. The binary index auto-refreshes when your data changes.

Two ways to consume this package

| Sub-path | Use it for | Loads WASM via | |---|---|---| | @sprigr/mcp-server (default) | Local Node.js stdio MCP server invoked by Claude Desktop / Cursor / etc. | --target nodejs build, auto-loads from filesystem | | @sprigr/mcp-server/runtime | Cloudflare Workers / bundled environments — embed the search engine directly in your Worker | Consumer injects a WASM module | | @sprigr/mcp-server/wasm-bundler | The companion --target bundler WASM build for the runtime sub-path | Imported as a module by your bundler | | @sprigr/mcp-server/wasm-nodejs | The --target nodejs WASM build, exposed for completeness | Auto-loads via fs |

Workers / bundler example:

import * as wasmModule from '@sprigr/mcp-server/wasm-bundler';
import { createSprigrRuntime } from '@sprigr/mcp-server/runtime';

const runtime = createSprigrRuntime({
  apiKey: env.SPRIGR_API_KEY,
  wasmModule,
});

const results = await runtime.search('my-index', 'wireless headphones');
//   ↑ runs locally via WASM (~5ms), or falls back to the cloud for semantic search

The runtime exposes search, importObjects, getObject, listIndexes, deleteObjects, and invalidate. It auto-invalidates the local engine after writes that go through it.

Quick Start

Add to your AI client config (Claude Desktop, Cursor, Windsurf, etc.):

{
  "mcpServers": {
    "sprigr": {
      "command": "npx",
      "args": ["-y", "@sprigr/mcp-server"],
      "env": {
        "SPRIGR_API_KEY": "your-api-key"
      }
    }
  }
}

Don't have an API key? Ask your AI agent to use the signup tool, or visit sprigr.com.

Environment Variables

| Variable | Required | Default | Description | |----------|----------|---------|-------------| | SPRIGR_API_KEY | Yes | — | Your Sprigr API key | | SPRIGR_API_BASE | No | https://search.sprigr.com | API base URL |

How It Works

AI Agent ──stdio──> @sprigr/mcp-server (local Node.js)
                         │
                         ├── search        → Local WASM engine (~5ms)
                         ├── import/delete → Sprigr API → invalidate local cache
                         └── On change    → re-download binary index → reload WASM

• Searches run entirely in-process via a compiled WASM search engine, with optional hybrid semantic search for improved relevance
• Writes (import, delete, create index, update settings) go to the Sprigr API
• After any write, the local binary index is automatically refreshed on the next search
• Cache version checks are throttled (5s TTL) to avoid unnecessary API calls

Available Tools

| Tool | Description | |------|-------------| | signup | Create a Sprigr account and get an API key | | list_indexes | List all indexes in your account | | create_index | Create a search index with searchable attributes and facets | | update_settings | Update index settings (searchable attributes, facets, synonyms) | | import_objects | Upsert JSON objects into an index (max 1000/call, same objectID = replace) | | search | Full-text search with filters, facets, pagination, and field selection | | multi_search | Search across multiple indexes in one call | | get_object | Fetch a single object by ID | | export_objects | Export all objects from an index as JSON (paginated) | | get_stats | Get account statistics | | delete_objects | Delete objects by ID (max 1000/call) | | delete_index | Permanently delete an entire index | | clear_index | Remove all objects but keep index settings | | set_alias | Point an alias name at an index (for zero-downtime reindexing) | | delete_alias | Remove an alias | | list_aliases | List all index aliases |

Requirements

• Node.js 18+

License

MIT