Azure AI Search MCP connector for Copilot Studio agents

March 23, 2026

Azure AI Search has 37 REST API operations across indexes, documents, indexers, data sources, skillsets, and synonym maps. Wrapping each one as a typed MCP tool would consume roughly 18,000 tokens on every tools/list call. This connector uses the Power Mission Control template to expose the entire API surface through three tools and eight MCP resources—about 1,500 tokens total.

What this connector does

The Azure AI Search MCP connector gives Copilot Studio agents full access to the Azure AI Search REST API (2025-09-01). Agents can search indexes, manage index definitions, run indexers, configure data sources, build skillsets, and monitor service health—all through progressive discovery.

Tools

The connector exposes three tools to the planner:

How it works

User: "Search my products index for wireless headphones"

1. Planner calls scan_search({query: "search documents"})
   → Returns: search_documents (POST /indexes/{indexName}/docs/search)

2. Planner calls launch_search({
     endpoint: "/indexes/products/docs/search",
     method: "POST",
     body: { search: "wireless headphones", queryType: "semantic" }
   })
   → Returns: matching documents with highlights and scores

The planner discovers the right operation first, then executes it with the correct endpoint and parameters. No upfront schema loading required.

Operations (37)

The capability index covers the full API surface:

Indexes

Search and documents

Indexers

Data sources

Skillsets

Synonym maps

Admin

MCP resources

Beyond tools, the connector exposes eight MCP resources for knowledge grounding. The planner can read these to understand what’s available before making tool calls.

Prerequisites

• Azure AI Search service (create one)
• Admin API key (full access) or query API key (search only)
• Power Platform environment with Copilot Studio

Setting up the connector

1. Create the custom connector

1. Go to make.powerapps.com > Custom connectors
2. Select New custom connector > Import an OpenAPI file
3. Upload apiDefinition.swagger.json
4. On the Code tab, paste the contents of script.csx
5. Save and test

2. Create a connection

When creating a connection, provide:

Get your keys from the Azure portal: Search service > Settings > Keys.

3. Add to Copilot Studio

1. Open your agent in Copilot Studio
2. Go to Tools > Add a tool > Custom connector
3. Select Azure AI Search MCP
4. The agent now has access to scan_search, launch_search, and sequence_search

Authentication

This connector uses API key authentication. The API key is stored as a connection parameter and injected as the api-key header on every request. Use an admin key for full access to all operations or a query key for read-only search and document retrieval.

Architecture

Copilot Studio Agent
    │
    ├─ tools/list → [scan_search, launch_search, sequence_search]
    │
    ├─ resources/list → [search://indexes, search://service/stats, ...]
    │   └─ resources/read → index schemas, stats, data source configs
    │
    ├─ scan_search({query: "search documents"})
    │   └─ Searches embedded capability index (37 operations)
    │
    ├─ launch_search({endpoint, method, body})
    │   ├─ Builds URL: {serviceUrl}/{endpoint}?api-version=2025-09-01
    │   ├─ Injects api-key header from connection parameter
    │   ├─ Handles 429 retry with Retry-After
    │   └─ Summarizes response (strip HTML, truncate)
    │
    └─ sequence_search({requests: [...]})
        └─ Executes requests sequentially, returns aggregated results

Resources

• Full source code on GitHub
• Azure AI Search REST API reference
• Azure AI Search data plane operations
• Power Mission Control template