m8ty_client_mcprag

Package overview

m8ty_client_mcprag solves this with embeddings and semantic tool search, allowing applications to discover relevant MCP tools from intent rather than rigid keyword or menu navigation.

MCP RAG client for semantic tool search, embeddings, and tool catalog access. Enables AI-powered discovery of API tools via vector similarity search.

Before you start

Check the documentation of the m8ty_client_mcprag library.

Import all needed libraries

import 'package:dio/dio.dart';
import 'package:m8ty_client_mcprag/m8ty_client_mcprag.dart';

Client Setup

final client = M8tyClientMcprag();
client.setOAuthToken('ApiOAuth2', accessToken);

final ragApi = client.getMcpRagApi();

Use basePathOverride for tests, staging, or tenant-specific routing:

final client = M8tyClientMcprag(
  basePathOverride: 'https://staging.example.com/api/v1',
);

API Surface

McpRagApi

Method	Description
`createMcpRagEmbedding`	Create text embeddings for semantic search
`getMcpRagCatalog`	Get the full MCP tool catalog
`getMcpRagVectorizedCatalog`	Get tool catalog with pre-computed vectors
`searchMcpRagTools`	Search tools by natural language or vector

How to perform different tasks

Here some examples how the dart client can be used

Create an embedding

final request = McpEmbedRequestModel((b) => b
  ..text = 'Find tools that can summarize PDF documents'
  ..inputType = 'query');

try {
  final response = await ragApi.createMcpRagEmbedding(
    mcpEmbedRequestModel: request,
  );

  final embedding = response.data;
  if (embedding == null) {
    throw StateError('Embedding response was empty.');
  }
  final vector = embedding.vector;
} on DioException catch (e) {
  throw StateError('Embedding request failed: ${e.response?.statusCode}');
}

Search tools by natural language

final request = McpToolSearchRequestModel((b) => b
  ..text = 'Find a tool that can retrieve a portfolio position'
  ..topK = 10
  ..minScore = 0.25);

final response = await ragApi.searchMcpRagTools(
  mcpToolSearchRequestModel: request,
);

final matches = response.data?.matches;
if (matches == null || matches.isEmpty) {
  return;
}

Search tools by precomputed vector

final request = McpToolSearchRequestModel((b) => b
  ..vector.addAll(queryVector)
  ..topK = 5);

final response = await ragApi.searchMcpRagTools(
  mcpToolSearchRequestModel: request,
);

Restrict search to specific tool names

final request = McpToolSearchRequestModel((b) => b
  ..text = 'Create an order'
  ..toolNames.addAll(['orders_create_order'])
  ..topK = 3);

final response = await ragApi.searchMcpRagTools(
  mcpToolSearchRequestModel: request,
);

Get tool catalog

final response = await ragApi.getMcpRagCatalog();
final catalog = response.data;

if (catalog == null) {
  return;
}

for (final tool in catalog.tools) {
  print('${tool.toolName}: ${tool.summary}');
}

Get vectorized catalog for synchronization

final response = await ragApi.getMcpRagVectorizedCatalog();
final catalog = response.data;

for (final item in catalog!.tools) {
  print('Tool: ${item.tool.toolName}, Vector length: ${item.vector.length}');
}

03 September 2026