LobeHub
LobeHub

Connectors

Connectors let agents do more than answer questions. They can use tools, follow specialized workflows, access services you authorize, and take action. Add task management, documents, or image generation; install community skills and MCP integrations; or connect GitHub, Notion, Google Drive, and other services.

Add the tools and skills your agent needs, then choose whether to pin them, use them automatically, or disable them. Connect and authorize an account before using an external service.

What can you connect?

TypePurposeExamples
LobeHub built-in capabilitiesManage tasks, create content, and work in LobeHubTask Tools, Documents, Message, Image Generation, Artifacts, and more
Third-party servicesAccess authorized resources and use exposed actionsGitHub, Linear, Gmail, Slack, and more
Community skillsSpecialized instructions, workflows, and supporting resourcesSkills tab in the Skill Store
Community MCP integrationsAccess external tools through the Model Context ProtocolMCP tab in the Skill Store
Your own skillsUse skills you created or obtained elsewhereGitHub, URL, and ZIP imports

Availability depends on the client, deployment settings, permissions, and execution environment. Installing a skill does not automatically grant access to external accounts or local files.

Use a connector

Open the Skills menu

  1. Open the Agent page.
  2. Click + at the lower-left corner of the chat input.
  3. Select Skills to view the agent’s pinned and automatically available tools and skills.
  4. Select Add Skill… to open the Skill Store, or Skills Marketplace to manage existing tools and skills.

The store has LobeHub, Skills, and MCP tabs. LobeHub offers built-in capabilities and service connectors; Skills and MCP offer community extensions. Use Add in the upper-right corner to import a skill from GitHub, a URL, or ZIP, or to add a custom MCP integration.

Add and authorize

  1. Find the item and review its purpose, dependencies, and required permissions.
  2. Add built-in capabilities as prompted. For an external service, select Connect and authorize an account.
  3. Install community extensions and configure required MCP endpoints, authentication, or script dependencies.
  4. Return to the agent’s skill list, check its usage mode, and request a task.

Installation, authorization, and activation for a particular agent are separate steps. A successful connection does not grant access to every resource.

Pinned, Auto, and Disabled

ModeMeaningBest for
PinnedLoad by default for this agent, subject to runtime and permission requirementsCore capabilities
AutoDiscover and activate when needed instead of loading everything on every turnOccasional use
DisabledExclude the capability from this agentCapabilities you do not want used

Choose a mode in the item’s action menu. The Auto group switch controls automatic activation, not bulk installation or removal. Unpinning is not uninstalling, and disabling does not revoke account authorization. Some system capabilities offer different options.

Then describe the work: “Save this research as a document” or “Find GitHub issues I can access.” For sending, modifying, or deleting, specify the target and allowed scope and inspect the result. Provide passwords and API keys through secure credential controls, not ordinary chat.

LobeHub built-in capabilities

LobeHub built-in capabilities cover task management, documents, messaging, image generation, calculation, and interactive content creation. Select the capabilities you need in the store’s LobeHub tab. Availability depends on your client and execution environment.

Task Tools

Purpose: Create and manage tasks, subtasks, dependencies, assignees, status, and comments through conversation. You can also set acceptance requirements and start execution. Your changes appear in the Task interface.

Try: “Create a launch task with copywriting, design, and final-review subtasks. Make the review depend on the first two.”

Check tool access and permissions before use. After creation, inspect the saved task, executing agent, inputs, and output requirements. To run recurring work, configure and enable a schedule, then check the timezone, next run, and status. A deadline alone does not trigger execution or notifications. See Scheduled Tasks. Before deletion, identify the target and whether subtasks are included.

Acceptance

Acceptance helps you assess whether the output meets your requirements. Configure acceptance criteria so the agent can check them before delivery. A failed check can hold back delivery and start a repair round. The report includes a verdict, score, and evidence for each check.

Use this workflow:

  1. Define checks before execution. Specify observable criteria, such as required sections, source links, time range, and an output that can be opened. Make each criterion more specific than "high quality."
  2. Confirm the configuration. Ask the Agent to add the checks and inspect the saved task. If acceptance checks are unavailable, use the criteria for manual review.
  3. Run and inspect. Review the deliverable and check report. Compare the evidence for each item with your requirements.
  4. Handle failures. Read the failed items and repair output. If work remains blocked or the repair is insufficient, supply missing information or request changes, then inspect the new result.
  5. Accept the result. Confirm completion through the available action when the result meets your requirements. Human acceptance is separate from the Acceptance report.

Example request:

Create a competitor report task and add acceptance checks: cover all three named companies; cite a source link and publication date for each pricing claim; state the observation period and currency; list unverified information separately. Show me the saved checks before execution.

Automated checks apply only to the requirements you configure. Review important facts and the final deliverable yourself, even when all checks pass.

Using the CLI

With the LobeHub CLI installed and authenticated to the intended deployment, the same task system is available through lh task:

*.sh
Shell
lh task list
lh task view T-1
lh task create -i "Draft the release notes"
lh task edit T-1 --status completed

Replace T-1 with the intended task identifier. Before marking a task complete, review its deliverable and check report. The last command only changes the status. Use your installed version's help for available options.

Documents

Purpose: Create and maintain the agent’s own persistent reports, notes, project context, and references.

With it enabled: Create, read, rename, copy, and delete documents; replace content or make targeted edits; list documents and adjust loading rules. Documents can belong to an agent or be associated with the current topic.

Try: “Create a competitor research document with sources. In future updates, change only pricing and product developments.”

Keep in mind: Uploaded source PDFs and images are available through the Resource Library or knowledge bases. See Agent Documents for the panel, scope, and history.

Message

Purpose: Work with configured messaging channels and manage their bots and connections.

With it enabled: Send, read, edit, and manage messages where the channel supports those actions; inspect or manage bot connections and routing.

Try: “Send the approved summary to our connected team channel. Let me review the text first.”

Keep in mind: Configure the platform and a valid connection first. Available history, editing, and deletion options vary by platform. Bot setup is separate from Slack and WhatsApp service connectors below; see Messaging Channels.

Image Generation

Purpose: Use image models configured in LobeHub, even when the current chat model does not produce images itself.

With it enabled: Discover available models and parameters, submit image-generation requests, and check status.

Try: “Generate a landscape product graphic with the subject on the right and headline space on the left.”

Keep in mind: Requires an available image model, provider configuration, and quota. Sizes and reference-image support depend on the model. After adding the tool, check that it is available in your conversation's runtime mode.

Credentials

Purpose: Help manage account authorization and sensitive credentials.

With it enabled: Initiate supported connection flows, guide secure credential storage, and supply environment variables in supported cloud execution environments.

Try: “Help me connect the service securely without pasting my key into chat.”

Keep in mind: Injecting credentials into a cloud sandbox does not provide them to a local device. Availability depends on where execution takes place.

Calculator

Purpose: Perform numerical calculations and symbolic mathematics.

With it enabled: Evaluate expressions, use variables, sort values, convert number bases, solve equations, and perform differentiation, integration, and limits.

Try: “Project revenue for the next year at 3% monthly growth and show each month.”

Keep in mind: Calculator does not retrieve external data. Use another tool to look up current exchange rates or prices first.

Artifacts

Purpose: Create and iterate on visual or interactive work in a dedicated preview panel.

With it enabled: Create HTML pages, React components, dashboards, SVG graphics, and simple browser games; revise colors, layout, or interactions through conversation.

Try: “Build a pricing page with monthly/yearly switching and show the preview, then add a dark mode.”

Keep in mind: Keep ordinary articles and code explanations in chat; use Documents for persistent text and Image Generation for model-generated images. You can prepare a self-contained preview without starting a cloud sandbox. External APIs, resources, and browser permissions may be restricted. General PDF/PPTX export is not part of this skill. Review the preview, test interactions, and use the available panel controls to inspect, copy, or save code.

LobeHub

Purpose: Guide the agent in operating platform resources through the LobeHub CLI.

With it enabled: Manage agents, topics, documents, knowledge bases, models, plugins, bots, and other supported resources when environment and permissions allow.

Try: “Rewrite this agent’s opening message for customer research.”

Keep in mind: Requires an available CLI and appropriate account permissions. Loading instructions does not bypass permissions or automatically perform changes.

Agent Browser

Purpose: Automate browser and Electron-page interactions.

With it enabled: Navigate, click, fill forms, capture screenshots, and extract content from sites requiring real-page interaction.

Try: “Open the product dashboard and summarize this month’s data. Let me sign in; do not submit changes.”

Keep in mind: Agent Browser requires an environment with device execution support and can be added from the desktop Skill Store. It differs from the runtime-managed Browser tool and web search; see Built-in Browser and Execution Devices.

Where are web search, memory, and knowledge bases?

Web search and memory have separate controls in the chat input’s + menu. Attach the relevant resources to use a knowledge base. Local system access and sandboxes depend on the selected execution environment. These capabilities do not always need to be added from the Skill Store.

Third-party service connectors

Connect the services below to let your agent work with authorized account content. After connecting, check the connector details for available actions and permissions. Actions may vary by account, service plan, and deployment configuration.

ServiceMain usesUsage notes
GitHubFind repositories, code, issues, and pull requests; support development collaborationRepository access and changes depend on your authorization. Confirm the target branch before merging a PR.
LinearWork with issues, projects, cycles, documents, and commentsAccess follows team and project permissions. Confirm the issue and fields before making changes.
Outlook CalendarView and manage calendar events and arrange meetingsConnect and authorize separately from Outlook Mail and OneDrive.
NotionWork with pages, databases, and knowledge contentGrant access to the pages and databases you need. Unshared content is not accessible.
PostHogQuery product analytics and inspect dashboards, experiments, and feature flagsSelect Connect in the Skill Store and follow the project authorization prompts.
XSearch posts, retrieve profiles, and support publishing and engagementAvailable actions depend on account permissions and the tools provided by the connector.
VercelInspect projects, deployments, logs, and deployment configurationCheck accessible projects after connecting. Confirm the target environment before changing deployment configuration.
GmailFind and read email; support composing, sending, and label managementReading, sending, and label management require the corresponding permissions.
Google DriveFind, organize, upload, and share filesAccess is limited to authorized files. Confirm recipients and access levels before sharing.
Google CalendarView calendars, create events, and arrange meetingsConfirm the calendar, timezone, and event-editing permissions.
SlackFind conversations, send messages, and work with channelsAccess is limited by channel permissions. Configure separately from bot connections in Message.
AirtableQuery, create, and update recordsAuthorize the bases and tables you need.
Google SheetsRead, write, and analyze spreadsheetsAuthorize the spreadsheet and specify the sheet and cell range before making changes.
Google DocsCreate, read, edit, and format documentsRequires access or editing permission for the target document.
SupabaseSupport database, authentication, and storage workflowsAuthorize the project and grant only the database and service permissions the task needs.
ConfluenceSearch pages and maintain knowledge content and space informationAccess is restricted by site, space, and page permissions.
JiraFind and create issues, update status, and track projects and sprintsCheck project permissions and allowed workflow transitions.
ClickUpCreate and update tasks and track projectsCheck workspace, list, and task permissions.
DropboxFind, upload, download, share, and organize filesConfirm the destination and access scope before sharing or moving files.
FigmaRead design information, inspect components, and export assetsRequires design-file access. See connector details for available actions.
HubSpotWork with contacts, deals, sales pipelines, and CRM workflowsContact, deal, and other data access follows account-role permissions.
OneDriveFind, organize, upload, download, and share filesConnect separately from Outlook Calendar and Outlook Mail; access is limited to authorized files.
Outlook MailFind, read, and send email and manage the inboxAuthorize separately from Outlook Calendar. Check recipients before sending.
SalesforceQuery CRM records, follow up leads, and update opportunitiesAccess is restricted by object, field, and organization permissions.
WhatsAppSupport messaging, media, and customer communication through the Business APIRequires a WhatsApp Business API account and follows business messaging rules.
YouTubeSearch videos and retrieve video information and available transcriptsTranscripts are available only when provided and accessible for the video.
ZendeskCreate, update, and track support tickets and retrieve customer informationAccess is restricted by support role and ticket permissions.
Cal.comCheck availability, book meetings, and manage event typesConfirm availability, timezone, and booking permissions.

Authorization and tool permissions

A successful connection does not grant unrestricted access. Individual tools can run automatically, require approval, or be disabled. Inspect the configuration: not every write operation is guaranteed to prompt for approval. Disabling a tool for an agent is separate from revoking the service authorization.

Community skills and MCP integrations

LobeHub’s community offers many third-party skills and MCP integrations. Go to Agent page → chat input + → Skills → Add Skill…, then use the Skills or MCP tab to find and add them.

  • Skills provide instructions, workflows, and references; some include scripts. Review their source, dependencies, and permissions before installation.
  • MCP integrations expose external tools and may require an endpoint, authentication, or execution environment. See MCP Marketplace.

Review the description and required permissions before installing, and use extensions from sources you trust. Configure account access or an execution environment for extensions that need external services or scripts.

Event subscriptions

Some connectors can tell an agent when something happens: a CI check fails, an issue is filed, a deployment finishes. Ask in a conversation, for example "Tell me when CI fails on this PR". The agent looks up the matching event on the connector and asks you to approve a subscription. Like any tool that needs approval, it follows your tool approval settings: with auto-run on, the subscription is created without asking. Subscriptions are made in a direct conversation with the agent; agents working in a Channel cannot subscribe.

The approval card shows the connector and event, the filter (which repository, team, or resource), and what the agent will do each time the event arrives. Each event runs the agent once in this topic and uses credits, and the subscription stays active until you delete it. To change the filter or the instruction, reject the request and tell the agent what you want instead.

After you approve, events arrive in the same topic as a message from the connector, with the event data and the instruction you approved. Events that arrive close together are handled in one run. The agent handles them without you; tools set to require approval still wait for you in the conversation. Event data comes from outside LobeHub, so the agent treats it as information, not as instructions.

A connector can be subscribed to when it connects to a remote MCP server that offers events, with an authorized account. Ask the agent what it can watch; it lists the events each connector offers.

Manage subscriptions

Open Settings → Automations, or the Automations tab in workspace settings. Each row shows the subscription, its status, the agent and topic it runs in, and when the last event arrived. To see only one agent's subscriptions, open the Automations tab of that agent's profile.

StatusMeaningWhat to do
ActiveEvents are deliveredNothing. A note under the status explains a skipped run.
Reconnect neededThe connector's authorization has expiredReconnect the connector in Settings → Connectors
Delivery failingThe connector reported delivery problems or paused deliveryClick Check
EndedThe connector ended the subscription, for example access revokedDelete it, then ask the agent to subscribe again if still needed

Check asks the connector whether the subscription is still delivering, and resumes delivery the connector has paused. Subscriptions are not checked automatically, so use Check when events you expect stop arriving.

Deleting a subscription stops its events from running the agent. Moving the topic or the agent to the trash, or deleting the connector, also removes its subscriptions; restoring the topic does not bring them back. In a workspace, every member can see the list, and only the member who approved a subscription or a workspace admin can check or delete it.

On a self-hosted deployment, an administrator must enable event subscriptions first. See Feature flags.

Import your own skills

Go to Agent page → chat input + → Skills → Add Skill…. In the store, use Add in the upper-right corner to import from GitHub, a URL, or ZIP.

Import from GitHub

  1. Choose Import from GitHub.
  2. Enter a public repository URL or the full directory URL for the skill.
  3. Confirm and review the parsed name, description, and resources.
  4. Return to the skill list and select Auto or Pinned as needed.

Choose a directory containing a valid SKILL.md. If the skill is in a subdirectory, use that directory's URL. Include the branch in the URL, especially if it is not main. If a link cannot be imported, package the skill as a ZIP and upload it instead.

Import from a URL

  1. Choose Import from URL.
  2. Enter a publicly accessible direct SKILL.md URL or a supported skill-package URL.
  3. Confirm, review the result, and choose a usage mode for the agent.

Use a direct link to the skill file or package, rather than its introduction page. LobeHub needs to retrieve and read the file, so private-network or protected addresses may not work.

Upload a ZIP file

  1. Package SKILL.md with the required references and scripts.
  2. Choose Upload Zip, then select or drop the file.
  3. Wait for upload and parsing, then review the result.
  4. Set the skill to Auto or Pinned for the current agent.

Place SKILL.md at the root or inside one top-level directory:

*.txt
Plaintext
my-skill/
├── SKILL.md
├── references/
└── scripts/

The UI also accepts .skill files, parsed as ZIP packages. Package skills individually; bulk import from deeply nested directories is not supported. Importing does not automatically install dependencies or grant scripts unrestricted execution access.

Frequently asked questions

Why is an installed capability not being used? Check whether it is disabled for this agent or automatic activation is off. Then check runtime mode, model tool-calling support, authorization, device connectivity, and dependencies. Pinning does not bypass those requirements.

Documents, Artifacts, or Image Generation? Use Documents for persistent reports and notes; Artifacts for web pages, interactive components, and SVG work; Image Generation for images produced by image models.

Why can’t I find a connector? Availability may vary by client, deployment settings, and account permissions. Check the entries shown in your Skill Store and confirm that your account can add connectors.