Connected Tools Checklist Before You Start a Task

Delegating a task to Claude that depends on outside data - a connector, a Skill, a Research mode query touching internal sources - works best when you've confirmed what's actually available first.

This checklist walks through what to verify before you start, so a task doesn't come back incomplete because a connector wasn't live or a Skill wasn't the one you thought it was.

How to Use This Checklist

Walk it before delegating any task that depends on a connector, Agent Skill, or Research mode query touching internal sources.
Skip the tiers that don't apply - a purely public-source research question doesn't need the connector items checked.
Revisit it periodically, not just once - access and available connectors change as a team's setup evolves.
Keep it short in practice: most of these are quick confirmations, not deep investigations.

Connectors and Data Access (1-8)

Identify which sources the task actually needs. List the specific systems - Google Drive, Slack, Notion, SharePoint, Confluence, GitHub, Jira, Salesforce - before checking anything else.
Confirm each needed connector is actually set up. A connector has to be explicitly configured; it isn't available by default just because your org uses that tool.
Confirm the connector covers the right scope. A Slack connector limited to one workspace or a Drive connector limited to one shared drive won't reach content outside that scope.
Check whether it's a first-party or MCP-based connector. The distinction affects what's supported and who maintains it if something looks off.
Verify your own account has access to the underlying content. A connector only surfaces what the connecting account can already see - it doesn't grant new access.
Check when the connector was last verified as working. A connector that silently stopped syncing can look "connected" in settings while returning stale or missing results.
Note anything the connector explicitly does not cover. A partial connector (one channel, one folder, one project) is easy to mistake for full coverage.
Confirm who can see the results if this task's output gets shared. Data pulled through a connector inherits real sensitivity - know who the output is safe to share with before you start.

Agent Skills (9-14)

Check whether a Skill already exists for this kind of task. A recurring task may already have a Skill built for it, worth using rather than re-explaining from scratch.
Confirm the Skill's description actually matches your task. A Skill with a close-but-not-exact description can either fail to trigger or trigger on the wrong task.
Decide whether to let Claude discover the Skill or invoke it by name. Naming a Skill directly is safer when several Skills could plausibly apply to the same request.
Verify the Skill has access to any tools it needs. A Skill scoped to specific tool access won't work if that access isn't also granted separately.
Check who built and last updated the Skill. A Skill that hasn't been touched since a process changed may produce output in an outdated format.
Test a new or recently changed Skill on a low-stakes example first. Don't trust an unfamiliar or recently edited Skill's first real run for something high-stakes.

Research Mode Readiness (15-20)

Decide whether the task needs public sources, internal sources, or both. This determines which connectors, if any, actually need to be checked beforehand.
Confirm connected sources are included in scope, not just available. Some setups require you to explicitly reference a connected source in your request rather than assuming Research mode will search it automatically.
Check whether the question is specific enough to investigate well. A vague question wastes a multi-step research run regardless of how well the connectors are set up.
Confirm you have time to wait for a multi-step run. Research mode takes longer than a normal reply - don't start one moments before you need the answer.
Plan for a review pass on the returned report. Budget time to check citations and source diversity before acting on or sharing the report.
Note any time-sensitive constraint up front. If the answer needs to reflect "as of today," say so explicitly in the question rather than assuming it's implied.

Team and Admin Considerations (21-25)

Confirm whether this connector or Skill is org-wide or personal. Personal connections don't extend to teammates who might rely on the same task later.
Check for an existing rollout or governance policy. Larger orgs often have a defined process for adding connectors - skipping it can create access gaps or duplicated setups.
Identify who to ask if a connector or Skill isn't working as expected. Know whether that's an IT admin, a Skill's original builder, or a broader support channel before you're blocked mid-task.
Confirm sensitive-data handling expectations for this task. Some connected sources carry compliance or governance requirements that affect what you're allowed to do with the output.
Re-check this list for any task that's meaningfully different from your last one. A checklist run for one task's connectors and scope doesn't automatically cover a different task's needs.

Applying the Checklist in Order

Items 1-8 (Connectors and Data Access) matter most - a task can't use a source that isn't actually connected, no matter how well everything else is set up.

Items 9-14 (Agent Skills) come next, since they shape how the task gets done once the data access question is settled.

Items 15-20 (Research Mode Readiness) apply specifically when the task involves a multi-step research query, and items 21-25 (Team and Admin Considerations) matter most in a team or enterprise setting rather than for individual, personal use.

FAQs

Do I need to run this whole checklist for every single task?

No - skip tiers that don't apply. A quick public-source question doesn't need the connector or admin items checked; a task pulling from Slack and Drive does.

What's the most common reason a delegated task comes back incomplete?

Assuming a connector is live when it isn't, or assuming it covers more scope (channels, folders, projects) than it actually does.

How do I know if a connector is first-party or MCP-based?

Check the connector's settings or documentation for that specific integration - the distinction is covered in more depth on this site's connectors and enterprise search page.

Should I always name an Agent Skill directly instead of letting Claude discover it?

Not always - automatic discovery works well most of the time. Naming a Skill directly is most useful when several Skills could plausibly apply to the same request and you want a specific one.

What should I do if I'm not sure whether my account has access to a connected source?

Ask directly, or try a small, low-stakes request against that source first - a connector only surfaces what the connecting account can already see, so unclear access on your end will surface as missing or partial results.

Is this checklist different for a personal account versus a team account?

The core items (1-20) apply either way; items 21-25 mostly apply once you're working in a team or enterprise setting with shared connectors and governance policies.

Why does connector scope matter more than whether a connector exists at all?

Because a connector that exists but covers the wrong folder, channel, or workspace can look "connected" while still missing the exact content a task needs - existence and coverage are two different questions.

What's the risk of skipping the review-pass item for Research Mode Readiness?

A returned report can look thorough while resting on weak or repetitive sources; skipping the planned review pass means that gap goes unnoticed before the report gets acted on or shared.

Do I need IT approval to use a connector someone else already set up?

Usually not for using an existing connector within your own access, but adding a new one at the team or org level typically goes through an admin decision - check your org's specific rollout policy.

How often should this checklist actually get revisited?

Whenever the task, the connected sources, or the available Skills have meaningfully changed since your last check - not on a fixed schedule, since access and setups evolve at their own pace.

Research Mode, Skills & Tools Basics - where checking connectors and Skills first appears as a basic habit
Connectors and Enterprise Search: Bringing Outside Data into Claude - the mechanics behind what this checklist verifies
What Agent Skills Are and How Claude Uses Them - background for the Agent Skills tier of this checklist
Connector Rollout Checklist for IT Admins - the admin-side counterpart to this checklist
Research Mode, Skills & Tools Best Practices - broader habits this checklist supports

Stack versions: Written against the Claude model lineup current as of ~June 2026 - Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5 (the default), and Claude Haiku 4.5. Model names, pricing, and product features move quickly - verify current specifics at platform.claude.com/docs before relying on them.