Copilot Connectors vs Power Platform Connectors in Copilot Studio

Use a synced Microsoft 365 Copilot connector when an agent needs to search a large body of external content with semantic indexing and citations. Use a Power Platform connector when the agent must query a system at runtime or perform an action such as creating, updating, or approving a record. Many production agents need both: indexed knowledge for discovery and a live connector for transactions.

This comparison uses “Copilot connector” to mean the synced connector described in Copilot Studio’s connector comparison. Microsoft 365 also supports federated, MCP-based Copilot connectors that retrieve content at query time; those are a separate real-time model and should not be confused with synced connectors.

Quick comparison

Criterion	Synced Copilot connector	Power Platform connector
Primary purpose	Ingest external content into Microsoft Graph for search and grounding.	Call an external API to retrieve current data or execute an operation.
Data access model	Content and metadata are copied into the Microsoft Graph index and synchronized.	Runtime request to the source system; real-time knowledge does not copy the business data into Microsoft 365.
Best content	Documents, knowledge articles, wikis, historical tickets, policies, and other searchable text.	Current order status, inventory, account records, approvals, case creation, and updates.
How Copilot Studio uses it	Knowledge source for retrieval and grounded answers.	Tool or action; supported connectors can also be added as real-time knowledge while that feature remains in preview.
Write operations	No; indexed content supports retrieval.	Yes, when the selected connector action and user permissions allow it.
Citations	Designed for search-grounded answers with references to indexed items.	Real-time knowledge can ground an answer; an action response does not automatically provide document-style citations.
Freshness	Depends on connector crawl and synchronization frequency.	Depends on the state and response of the source API at request time.
Latency	Retrieval is served from the index after ingestion.	Includes a live round trip to the target service and may be affected by throttling or outages.
Permissions	Indexed items carry access control information from the source.	Runtime calls use the configured connection and authentication mode; real-time knowledge calls use the user’s identity.
Administration	Tenant administrators configure connectors, schema, semantic properties, ACLs, and synchronization.	Makers configure connections and operations in a Power Platform environment; administrators govern them with data policies.

Mental model: library index vs live service desk

A synced Copilot connector works like a library index. It imports external items into Microsoft Graph, enriches them with metadata and access control lists, and makes the content searchable across supported Microsoft 365 experiences. The source remains authoritative, but answers are retrieved from the synchronized index.

A Power Platform connector works like a live service desk. The agent sends a request to an API when the user asks a question or invokes a task. It can read the current value and, when an action is configured, change the source system.

The distinction affects architecture:

• Index then answer: best for finding and synthesizing relevant information across a large corpus.
• Call then answer or act: best for current values and transactions.

When to choose a synced Copilot connector

Choose the indexed approach when most of these statements are true:

• Users ask broad questions across many documents or knowledge articles.
• Semantic ranking is more useful than querying one known record by ID.
• The same external content should appear in Microsoft Search or other supported Microsoft 365 experiences.
• Answers should cite the external items used for grounding.
• Short synchronization delay is acceptable.
• The source permissions can be represented accurately as item ACLs.

Typical sources include external wikis, service-management knowledge bases, issue repositories, document stores, and engineering project content. Microsoft maintains a gallery of supported connectors, and tenant administrators can also implement custom ingestion through the Copilot connectors API.

When to choose a Power Platform connector

Choose a Power Platform connector when one or more of these requirements applies:

• The answer must reflect the source system’s current state.
• Data should remain in the source rather than being copied into the Graph index.
• The agent must create, update, approve, delete, or submit something.
• The request targets a specific entity, such as an incident, account, order, asset, or invoice.
• The API already exposes the required operation and authentication model.
• A prebuilt connector exists, or your team can define a custom connector for the API.

Power Platform connectors can be added to an agent as tools at the agent level or within a topic. Supported connectors can also provide real-time knowledge, but Microsoft currently labels that Copilot Studio capability as preview. Preview features can change and may not meet production support requirements.

Important naming change: Copilot connectors can also be federated

Microsoft 365 Copilot connectors now include two models:

• Synced connectors ingest and index content in Microsoft Graph.
• Federated connectors retrieve content at query time through MCP without indexing it into Microsoft Graph.

The official Copilot Studio comparison still contrasts synced Copilot connectors with Power Platform connectors. When reviewing an architecture proposal, ask whether “Copilot connector” means synced or federated. The data-movement, schema, latency, and availability answers differ.

Security and identity

Synced Copilot connectors

Each indexed external item includes metadata and an access control list. Search and Copilot should return an item only to a user represented in that ACL. This makes correct identity mapping and permission synchronization essential. A connector that indexes content correctly but assigns the wrong ACL can expose or hide information.

Power Platform connectors

Authentication depends on the connector and connection configuration. For real-time knowledge, Microsoft states that calls run at request time with the user’s authentication token, preserving source-system access control. Tools and actions can use supported user- or maker-provided connection patterns, depending on the scenario.

Power Platform data policies govern which connectors can be used together and where data may flow. Treat data policy, connector credentials, environment strategy, and least-privilege API permissions as part of the design—not as final deployment tasks.

Architecture patterns

Pattern 1: searchable knowledge only

1. An administrator configures a synced Copilot connector.
2. The connector maps content, metadata, semantic properties, and ACLs.
3. External items are synchronized into Microsoft Graph.
4. The Copilot Studio agent uses the configured enterprise connection as knowledge.
5. User questions retrieve indexed items, and the agent produces a grounded answer with citations.

Pattern 2: live lookup or transaction

1. A maker adds a prebuilt or custom Power Platform connector.
2. The required operation is exposed as an agent tool or action.
3. The user asks for a current value or requests a change.
4. The agent calls the source API under the configured identity.
5. The source returns data or confirms the operation, and the agent responds.

Pattern 3: hybrid knowledge and action

Use an indexed connector to explain the policy and a Power Platform connector to carry out the approved task. For example, an IT agent can cite troubleshooting and eligibility articles from an indexed knowledge base, retrieve the current incident status through a live connector, and create an escalation through an action.

Practical decision questions

1. Is the user searching or transacting? Search favors indexing; transactions require a live action.
2. How fresh must the answer be? If minutes matter, check sync frequency or use runtime retrieval.
3. May the content be copied? Regulatory or residency requirements may rule out indexing.
4. How large is the corpus? Large unstructured collections benefit from semantic indexing; one known record is better fetched directly.
5. Should other Microsoft 365 experiences reuse the content? Synced connectors can also power supported Microsoft Search and Copilot experiences.
6. What happens if the source is unavailable? Indexed retrieval may continue with the last synchronized content; a live call may fail or time out.
7. Which identity should perform a write? Define user consent, maker-provided credentials, approvals, and audit requirements before enabling actions.

Setup checklist for a synced Copilot connector

• Select a Microsoft-provided connector or design a custom ingestion connector.
• Map the schema, searchable content, retrievable properties, and semantic properties.
• Map users and groups accurately for source ACL enforcement.
• Configure synchronization and monitor ingestion errors, deletions, and crawl freshness.
• In Copilot Studio, add the tenant-configured connector as an enterprise knowledge source.
• Test with users who have different source permissions.
• Open citations and confirm that they resolve to accessible source items.

Setup checklist for a Power Platform connector

• Choose the Power Platform environment and review its data policies.
• Select a prebuilt connector or create a custom connector from a documented API.
• Configure the minimum authentication scopes required.
• Add specific operations as tools or actions; avoid exposing an entire API without need.
• For real-time knowledge, confirm that the connector is supported and that preview use is acceptable.
• Test consent and connection behavior in every publishing channel.
• Handle timeouts, throttling, missing records, validation errors, and partial failures.
• Log important write operations and require confirmation for consequential actions.

Common mistakes

• Indexing rapidly changing transactional data: search results can be stale between synchronization cycles.
• Using a live API for broad document search: the source API may lack semantic ranking, citation metadata, or efficient full-text retrieval.
• Assuming every Copilot connector is indexed: federated Copilot connectors are a different query-time model.
• Treating preview knowledge as a guaranteed production feature: review availability, limits, and support status before deployment.
• Testing only as an administrator: verify least-privileged users and users with no source access.
• Ignoring connector error states: monitor crawl failures for indexed sources and runtime API failures for live sources.
• Letting an agent write without confirmation: add validation, clear user confirmation, and audit logging for material changes.

Recommended default

Start with a synced Copilot connector for stable, searchable knowledge that should be reused across Microsoft 365. Add a Power Platform connector for live fields and operations. If copying content is prohibited, evaluate Power Platform real-time knowledge or a supported federated Copilot connector, then confirm preview status and production requirements in the current Microsoft comparison.