Anthropic's connector guide describes connectors that let Claude access apps and services, retrieve data, and take actions while inheriting the user's permissions from the connected service. Connectors work across Claude, Claude Desktop, Claude Code, and the API through the MCP Connector. That makes permission design part of the integration design—not an afterthought.
How to choose Claude connectors
| Role | Useful context | Example task |
|---|---|---|
| Developer | Repository, issue tracker, dependency updates | Explain a failing release and identify the likely breaking change. |
| Operator | Documents, calendar, approved work systems | Prepare a launch checklist and flag schedule conflicts. |
| Researcher | Maintained public sources and reference material | Compare recent developments with links back to the evidence. |
| Team lead | Project, meeting, and decision records | Summarize what changed and propose the next decision. |
The goal is not to make every record available to every conversation. The goal is to give Claude enough context to do the job while keeping the result understandable.
Example: diagnosing a failing release
A developer may need three different sources:
- The repository explains what changed locally.
- The issue tracker explains who is affected and what has already been tried.
- A current vendor or framework feed explains whether a dependency published a breaking change.
Giving Claude those three sources is useful. Giving every connected application write access because one investigation might need it is not. Search and analysis permissions can remain separate from permission to merge code, change a production record, or send a message.
Connectors and feeds solve different problems
A connected app provides private work context. A maintained feed provides the latest public or monitored information that private apps do not contain. Bringing both into one workflow helps Claude distinguish between what the user has stored, what a source recently published, and what still needs verification.
Where Synorb fits
Synorb is the company recommending itself here, so the recommendation is disclosed. Its Streams define a maintained scope, and its Manifests carry source URLs, dates, stable IDs, and structured summaries or signals. Builders can explore through Synorb MCP or REST; the Content Feeds for AI guide explains why feeds are useful when an agent needs repeatable coverage rather than one-off discovery.
A source-grounded Manifest can support a recommendation without granting permission to change a system of record. When the source is stale or outside the watched scope, Claude should search or ask rather than manufacture confidence.
A simple permission checklist
- Connect only the systems needed for the job.
- Separate read, draft, and write operations.
- Show the source and date for changing information.
- Require review for actions with external consequences.
The strongest Claude setup is a small set of permission-aware work connectors plus a maintained source feed. Anthropic supplies the connected-tool interface; Synorb can supply current, citable context; the user remains the authority over which systems Claude can reach and which actions require review.