Describes connectors and plugins that link the Loop to external tools, MCP standard, and security best practices for permissions and secret management
On its own, a loop can only touch the files in front of it. Real work usually lives in external tools — for example, GitHub, Slack, or project trackers. Connectors and plugins are how the loop reaches those tools and participates in real work.
Connectors give the Loop a single, focused link to an external tool. Plugins are installable bundles that can contain multiple connectors plus skills and tools. Use connectors for targeted access; use plugins to install a package of capabilities in one step.
ConnectorsA connector links the Loop to one external tool. Picture a cable running from the Loop to GitHub, Slack, or a project tracker. Once that link exists, the Loop can read and act in that tool — for example, it can read an open issue, post a message, or check task status.
The connector is the bridge that enables those actions.PluginsA plugin is an installable bundle. Inside a plugin there can be multiple skills, tools, and connectors packaged together. The distinction is one of scope: a connector is a single link to one tool (one cable), while a plugin is a kit that can include many pieces and be installed in a single step.
MCP — Model Context ProtocolSome connectors follow an open (or emerging) standard called MCP — the Model Context Protocol. MCP defines a shared set of rules and message formats for how an AI agent connects to an external tool. In short, it’s a common language for communication between agents and tools.
A helpful analogy is a standard plug, like USB: before a common plug, every device required a unique cable. With a shared shape, any device fits any port. MCP does the same for AI tools: a toolmaker implements one MCP connector, and any agent that speaks MCP can use it. Because of this shared standard, connectors built for one tool can often be adapted to others with little extra effort.Why connectors and plugins matterConnectors and plugins give the Loop hands. Without them, the Loop is limited to local files. With connectors and plugins, the Loop can:
Pull an issue from a tracker.
Open a pull request on GitHub.
Send updates to a team chat.
This reach is what lets the Loop participate in real work rather than only edit files locally. More reach, however, increases responsibility.Security and trust considerationsOnly add the connectors you actually need. Each extra link increases the potential surface for errors or misuse. Be deliberate about permissions and secret management.Bad practice example — storing secrets in source code:
# Bad: storing secrets in source code (do not do this)API_KEY = "sk-live-REDACTED"
Never commit real secrets (API keys, passwords) to source control or paste them into prompts. Use secure credential stores, environment variables, or secret managers instead.
Also be sure you know precisely what a connector can do before enabling it. Typical connector privileges vary; understand whether a connector is read-only, can change specific items, or has full access.
Connector privileges — typical breakdown
Privilege Level
Typical Actions
Example / Mitigation
Read-only
View issues, comments, or metadata
Use for analytics or triage tools; restrict to minimal scopes
Change (limited)
Post comments, update status, create branches
Grant only to specific repos/projects; use scoped tokens
Full access
Merge PRs, delete repos, manage permissions
Avoid unless absolutely necessary; require approval workflows and auditing
Best practices checklist
Add only necessary connectors and plugins.
Use short-lived or scoped credentials and secret managers.
Review and limit permissions; apply least privilege.
Audit connector activity and require approvals for high-risk actions.
Document which connectors are enabled and why.
Recap
A connector links the Loop to a single external tool.
A plugin is a package that can include skills, tools, and connectors, installed in one step.
Both extend the Loop’s reach beyond local files.
Trust matters: add only necessary connectors, keep secrets out of code and prompts, and verify connector permissions before enabling them.