Webex Connect AI Skills and Plugin

Install the complete bundle from the public Webex Skills repository.
The Webex Connect AI plugin package contains the skills and supporting components needed by plugin hosts. The skill catalog links to all 19 workflows and their individual packages.

What you are installing

The repository provides two distribution options:

• Standalone skills: individual SKILL.md workflows that can be copied into an AI tool's skills directory. Use this option when you want only a specific workflow, such as SMS or a live-chat widget.
• Webex Connect AI plugin: a packaged collection of skills, commands, rules, manifests, and branding. Use this option when you want the complete Webex Connect workflow set.

Installing the plugin does not automatically create a Webex Connect tenant, register a mobile device, verify a sender, or configure an MCP connection. Those are separate setup steps.

Prerequisites

Before installing, make sure you have:

1. An AI coding tool that supports the distribution you plan to use: Codex, Claude Code, or Cursor for the plugin. The standalone installers also support GitHub Copilot.
2. Git and network access to github.com/webex/webex-skills for the checkout-based installation instructions below. Individual skill installers also require Node.js 20 or newer and npm (which provides npx). The setup utilities require Python 3. The shell examples use macOS/Linux syntax; use the corresponding paths and copy commands on Windows.
3. A Webex Connect tenant for API-backed work. You will need to know whether the request is for sandbox or production before running setup.
4. The credentials required by the selected workflow:
   • Messaging, Event, Voice, and inbound webhook APIs: Service Key or service-specific JWT.
   • Profile and Contact Policy APIs: Profile Key.
   • Digital Asset API: Webex personal access token, organization ID, and the Digital Channel Onboarding API host.
   • SMS/Email MCP tools: OAuth 2.0 access token with spark:mcp, configured in the selected MCP host. This authentication is separate from direct APIs.
   • Webhook signature verification: configured Hub Secret.
5. Channel prerequisites where applicable:
   • SMS or email: an approved sender ID and recipient details.
   • Push: a configured mobile or web app asset and a registered user or device identifier.
   • Live Chat/In-App: a configured asset, app/user identifiers, and any required flow or webhook configuration.
   • Android or iOS SDK work: the relevant SDK, platform credentials, and a test application or device.
6. Network access to the AI tool, Webex Connect regional APIs, and any documentation or package registries required by the selected workflow.

Do not place service keys, JWTs, MCP OAuth credentials, APNs credentials, FCM credentials, or tenant secrets in this repository, a prompt, a committed .env file, or a public issue. Setup creates a local ignored .env; users enter secret values locally and restrict the file to their account.

Choose skills or the plugin

Use standalone skills when you want a small, auditable installation. Use the plugin when you want the full set of Webex Connect workflows and the bundled commands and rules.

The plugin contains skills for tenant setup and Chat, Email, Facebook Messenger, and WhatsApp digital-asset create, retrieve, and update operations, plus SMS, MMS, RCS, email, push, live chat, WhatsApp, Facebook Messenger, Apple Messages for Business, voice, webhooks, Contact Policy, sandbox testing, and Android or iOS Push and In-App Messaging builders.

The outbound messaging router is webexconnect-sendmessage. It selects one channel-specific adapter and adds shared setup, Contact Policy, or webhook guidance only when needed.

Skill catalog

The plugin contains the following 19 skills. Each skill is also linked from the public catalog. Each skill is intentionally narrow: use the router for a send request, then use the selected channel skill for the API-specific payload and rules.

Install the bundled plugin

Clone the public repository once. Use a directory where you keep development tools, rather than your application project's root:

git clone https://github.com/webex/webex-skills.git
cd webex-skills

The checkout already contains the assembled plugin at plugin-packages/webex-connect-ai/; no local build is required to install it. The plugins/webex-connect-ai/ directory is the source recipe used by maintainers, so use the assembled package for local plugin loading.

Codex

From the webex-skills checkout, add the marketplace and install the bundle:

codex plugin marketplace add .
codex plugin add webex-connect-ai@webex-skills

Alternatively, register the public Git repository directly:

codex plugin marketplace add https://github.com/webex/webex-skills.git
codex plugin add webex-connect-ai@webex-skills

Choose one marketplace source. The marketplace name is webex-skills, and the plugin name is webex-connect-ai.

Restart the Codex desktop app after installation, then verify that Webex Connect AI appears in the installed plugins list. Codex marketplace sources can be inspected or refreshed with:

codex plugin marketplace list
codex plugin marketplace upgrade

Claude Code

In a Claude Code session, add the public marketplace and install the bundle:

/plugin marketplace add webex/webex-skills
/plugin install webex-connect-ai@webex-skills

Review the plugin details and choose user, project, or local scope when the host prompts. For a local session using the cloned package directly:

claude --plugin-dir ./plugin-packages/webex-connect-ai

Start a new Claude Code session after installation. You can also run /plugin to inspect installed, enabled, and available plugins.

Cursor

From the webex-skills checkout, copy the assembled plugin into Cursor's local plugin directory:

mkdir -p ~/.cursor/plugins/local/webex-connect-ai
cp -R ./plugin-packages/webex-connect-ai/. ~/.cursor/plugins/local/webex-connect-ai/

Reload Cursor, open Customize, and confirm that the skills, commands, and rules are visible. Enterprise administrators may need to allow local plugin imports or distribute the plugin through a team marketplace.

For managed team distribution, an administrator can import the repository's
Cursor marketplace using Cursor's team marketplace controls. A public GitHub package does not by itself imply a listing in Cursor's official marketplace. See the
Cursor plugin installation guide for local imports and team marketplace requirements.

Install individual standalone skills

Choose a workflow from the skill catalog and follow its package README. Individual packages are distributed as GitHub archives; the commands below use npx to run their installers and do not require an npm registry publication.

For example, install the SMS skill for Codex at user scope:

npx --yes \
  --package https://raw.githubusercontent.com/webex/webex-skills/main/skills/webexconnect-sms-sendmessage/release/webexconnect-sms-sendmessage-latest.tgz \
  webexconnect-sms-sendmessage install --client codex --scope user

For API-backed work, also install the setup skill if it is not already available:

npx --yes \
  --package https://raw.githubusercontent.com/webex/webex-skills/main/skills/webex-connect-setup/release/webex-connect-setup-latest.tgz \
  webex-connect-setup install --client codex --scope user

Use --scope project from your application project to install there instead.
Use --client claude, --client copilot, or --client all for the other
installer-supported clients. all installs for Codex, GitHub Copilot, and Claude Code. See the SMS package README and setup package README for their installation options and release links.

To pin a standalone version, replace latest in its archive URL with an existing version from that skill's release/ directory. Avoid installing the same workflow both as a standalone skill and through the plugin, because the host may discover duplicate versions.

Standalone skills do not include the plugin's marketplace metadata, commands, branding, or MCP configuration.

Update an existing installation

Check the public plugin catalog for the bundle version, and the plugin release directory for available release archives and changelog. The installable checkout package is under plugin-packages/webex-connect-ai/. Plugin release archives, when published, are under plugins/webex-connect-ai/release/. Updating the plugin or skill files does not change the customer's Webex Connect tenant, digital assets, sender IDs, project .env, or securely stored MCP credentials. Do not delete those values as part of an update.

Before updating, close active agent sessions that are using Webex Connect AI. After the update, restart or reload the host so it does not keep an older skill copy in memory.

Codex plugin

For a Git-backed marketplace, refresh the source and inspect the installed plugin:

codex plugin marketplace list
codex plugin marketplace upgrade
codex plugin list

If the host still shows an older installed copy, use the plugin's update action in the installed-plugin view and restart Codex. Marketplace refresh and installed-plugin update may be separate actions in your host version.

For a local checkout, pull the latest source from the webex-skills directory:

git pull --ff-only
codex plugin marketplace add .

Run this on an unmodified checkout. If Git reports local changes or diverging history, keep those changes and resolve them before updating. Refresh the existing marketplace source and verify the installed version after restarting.

Claude Code plugin

Update the marketplace source and then update the installed plugin:

/plugin marketplace update webex-skills

Then update Webex Connect AI from the Installed tab of /plugin, or use the shell commands:

claude plugin marketplace update webex-skills
claude plugin update webex-connect-ai@webex-skills
claude plugin list

Start a new Claude Code session, or use the host's plugin reload action, after the update. A local --plugin-dir checkout is not updated by the marketplace; replace that checkout with the approved release contents and start Claude Code again.

Cursor plugin

For a marketplace installation, open Customize, locate Webex Connect AI, and select Update when available. If the plugin comes from a team or Git marketplace, the marketplace administrator must refresh the marketplace source first. Reload the Cursor window after the update and confirm that the new version and expected skills, commands, and rules are visible.

For a local installation, replace the contents of the local plugin directory with the updated checkout's plugin-packages/webex-connect-ai/ directory, then reload Cursor. Do not edit the installed plugin's generated copy directly; make changes in the canonical release source instead.

Standalone skills

Review the new version in the selected skill's package README and release directory. Rerun its installer with the same client and scope and add --force to replace that installed skill. For example:

npx --yes \
  --package https://raw.githubusercontent.com/webex/webex-skills/main/skills/webexconnect-sms-sendmessage/release/webexconnect-sms-sendmessage-latest.tgz \
  webexconnect-sms-sendmessage install --client codex --scope user --force

For a project installation, run from the target application project and use --scope project. Inspect any local modifications before replacing a skill. Update the setup skill and other installed Webex Connect dependencies as needed, preserving unrelated skills and your application's .env.

Verify the update

After updating any distribution:

1. Confirm the installed source or plugin reports the intended release tag.
2. Reload or restart the AI host.
3. Ask the agent to list the Webex Connect skills and confirm that the expected catalog is present.
4. Run webex-connect-setup for a preflight check if API-backed work is planned. Setup preserves existing .env values and only adds missing configuration keys.
5. Re-authenticate SMS or Email MCP access only if the host reports that its secure credential connection is missing or expired. Never paste a bearer token into the prompt, skill files, plugin files, or a committed .env.

If an update appears incomplete, inspect the host's installed-plugin list and source path before reinstalling. Do not mix skill directories from different release tags, because that can create mismatched API instructions and plugin
metadata.

Configure Webex Connect

After installation, open your application project in the AI host. Run setup there so that its local .env belongs to that application. Before sending messages, creating assets, or calling other APIs, run setup when tenant, region, authentication, or channel configuration is missing:

Use webex-connect-setup to configure Webex Connect.

Android, iOS, and browser widget builders collect their SDK configuration directly and do not require the API .env preflight.

The setup flow follows this order:

1. Select the environment. Explicitly choose sandbox or production. Do not allow the agent to silently default to sandbox or production.
2. Provide tenant context and confirm the region. Supply the tenant URL and organization ID when the selected APIs require them. If the region can be inferred from the tenant domain, show the inferred value and ask the user to confirm it. Resolve separate API hosts from the current official endpoint references; do not derive one API family from another.
3. Register Production MCP automatically. For Production, select the region-matched MCP URL from the table below and register it with the active AI host as part of setup. Sandbox has no MCP endpoint and must not configure the Production MCP server.
4. Create or update .env. Run the setup skill's bundled scripts/ensure_env.py with the selected environment and confirmed region. It automatically writes the supported Messaging v1/v2, Voice, Contact Policy, Profile, Audit Logs, Digital Asset, and Production MCP endpoints. It preserves existing values, adds missing keys from the blank template, sets owner-only file permissions, and leaves missing credentials blank for local entry. Do not ask the user to manually derive each endpoint.
5. Configure the needed authentication once. Messaging/Event/Voice and inbound webhook APIs use Service Key or service JWT; Profile use Profile Key; Digital Assets use a Webex personal access token plus organization ID. SMS/Email MCP tools use one bearer token with spark:mcp, separate from direct API credentials.
6. Check the file before API use. API-backed skills require the project .env and their specific endpoint/credential values. If either is missing, they stop and send the user back to setup. Android, iOS, and browser widget builders are exempt because they use SDK configuration.
7. Configure channel prerequisites. Confirm the sender, app asset, user registration, flow, or webhook required by the requested operation.
8. Review before mutation. The agent should show a redacted request summary and receive confirmation before creating an asset or sending a message.
9. Interpret the response correctly. An accepted or queued response means the platform accepted the request. It does not prove delivery, display, or user engagement.

To initialize the file manually from a project directory:

python3 /path/to/webex-connect-setup/scripts/ensure_env.py \
  --env-file .env --environment production --region in \
  --set WEBEX_CONNECT_TENANT_DOMAIN=tenantname.in.webexconnect.io \
  --set WEBEX_CONNECT_ORGANIZATION_ID=YOUR_ORGANIZATION_ID

The project .env is not automatically loaded by every MCP host. Setup registers the regional MCP URL automatically for Production; configure only the bearer token once through the selected host's supported secure credential flow or documented environment-variable wiring. Never paste MCP tokens into chat or assume writing .env alone authenticates a remote MCP.

Production MCP region mapping

The setup workflow asks for the tenant region before it configures MCP. Use
only these documented Production endpoints:

If the tenant region is not listed, do not guess an endpoint. Ask the user to verify the region against the Connect CPaaS MCP Server documentation and use direct APIs until a supported MCP endpoint is confirmed.

The MCP server must be enabled by an organization administrator in Webex Control Hub and in Webex Connect Tenant Settings. An approved SMS phone number or supported email asset and an approved AI Agent Messaging use case are also required. The setup flow registers the endpoint after Production and region selection, then asks only for the bearer token. Credentials are entered through the selected host's secure configuration and are never bundled with the skill or plugin.

The setup skill registers the selected regional endpoint with the active host as part of Production setup, before asking for authentication. For example, for a Production Mumbai tenant using Codex:

python3 /path/to/webex-connect-setup/scripts/install_mcp.py \
  --environment production --region in --client codex

Use --client claude or --client cursor for the corresponding host. The installer writes only the MCP URL, refuses Sandbox, and does not accept or store a bearer token. After it finishes, add only the bearer token through the host's secure MCP authentication flow and verify the connection. Do not provide MCP client ID or client secret values unless the selected host explicitly requires them.

For host-specific connection details, use the official MCP client guides:

• Codex: add the selected URL to ~/.codex/config.toml and provide the Webex OAuth 2.0 bearer token with the spark:mcp scope through the host's secure configuration.
• Claude Code: add the selected URL with claude mcp add --transport http and complete the browser OAuth flow from /mcp.
• Cursor: use the MCP install flow in Customize with the selected URL and Webex OAuth 2.0 bearer token with the spark:mcp scope, or configure the selected URL in ~/.cursor/mcp.json with Webex OAuth credentials.

The MCP server exposes sender discovery plus non-templated SMS and Email messaging. The documented tools are webex-sender-ids, webex-sms-message, and webex-email-message.

Sandbox behavior

Sandbox is intended for prototyping and supported test channels. The current Webex Connect workflow does not configure or use the production CPaaS MCP server for sandbox setup. Do not configure the Production MCP server or its OAuth credentials for a sandbox-only setup.

Use the sandbox skill for sandbox-specific limits and supported channels:

Use webexconnect-sandbox to plan a sandbox SMS, Voice, or WhatsApp test.

Production behavior

Production setup supports separate direct API and MCP authentication paths:

• Direct API path: configure the tenant region and the credential required by the API family in .env.
• MCP path: for SMS, Email, or sender discovery, use an OAuth 2.0 access token generated from an Integration App with the spark:mcp scope. The MCP client handles authentication; .env is not automatically passed to it by all hosts. Store secrets only in local ignored files or the host's secure credential store, never in the skill or plugin package.

MCP and direct API credentials are not interchangeable. A message response alone does not identify the tenant URL or environment, so retain the selected tenant and environment in the local setup record without storing credential values in that record.

End-to-end usage flow

After installation and setup, use natural-language requests. The agent should load the smallest relevant skill, collect missing inputs, validate the request, and ask for confirmation before an external write.

Example: send an SMS

Use webexconnect-sms-sendmessage to send this approved utility SMS to
+14155550123 using the configured sender ID: Your order has shipped.

The workflow validates E.164 formatting, sender availability, encoding, and the Send Message API v2 payload. It reports the acceptance status and message ID without claiming delivery.

Example: send an email through MCP

Use the Webex Connect email skill. Discover the approved marketing sender,
show me the sender and recipient, then send the confirmed campaign email.

For a production MCP workflow, setup must already have MCP access. The agent should not invent a sender ID when discovery does not return a compatible one.

Example: manage a digital asset

For creation:

Create a Webex Connect digital asset.

For retrieval or update, name the operation and provide the documented asset identifier:

Get the Chat digital asset with ID 12345.
Update the description of the Email digital asset with ID 12346.

The digital-asset workflow asks whether the operation is Create, Get, or Update, then asks whether the asset is Chat, Email, Facebook Messenger, or WhatsApp. It loads only the selected operation's documented schema, resolves the tenant route, and asks for confirmation before a create or update mutation.

Example: build a live-chat widget

Use webexconnect-livechat-widget to build a browser live-chat widget for my
Webex Connect tenant.

The workflow collects the tenant and asset details, uses the current Webex Connect JavaScript SDK guidance, and produces a runnable implementation with the required event and error handling.

Example: build a mobile app integration

Use webexconnect-ios-push-inapp-messaging-builder to build an iOS Push and
In-App Messaging demo.

For Android, use webexconnect-android-push-inapp-messaging-builder. The builder workflows collect app identifiers and platform credentials, generate the integration structure, and include validation and QA checks.

Common customer use cases

Transactional and marketing messaging

Use SMS, email, MMS, RCS, WhatsApp, or push skills to create approved outbound campaigns, order updates, offers, reminders, and service notifications. Use Contact Policy when consent groups, opt-out rules, or frequency caps apply.

Digital-channel onboarding

Create, retrieve, and update Chat, Email, Facebook Messenger, or WhatsApp digital assets interactively, associate them with the appropriate tenant configuration, and validate the request against the selected operation schema.

Customer support and live engagement

Build a browser live-chat widget or use Live Chat/In-App Messaging APIs for threaded conversations, attachments, quick replies, and notifications.

Mobile engagement

Build Android or iOS applications that register users and devices, receive push notifications, display rich content, and exchange in-app messages.

Delivery operations and observability

Use webhook and event skills to process delivery receipts, inbound messages, failures, and status callbacks. Treat API acceptance, provider delivery, and end-user display as separate states.

Prototyping before production

Use sandbox guidance to test supported channels and payloads. Move to production only after tenant access, sender approval, consent, credentials, callbacks, and channel assets are ready.

Security and operational guidance

• Review the plugin source and manifest before installation, especially when installing from a third-party marketplace.
• Grant MCP and filesystem permissions only to the workflows that need them.
• Keep secrets in the ignored local .env with owner-only permissions or in the host's secure credential store. The MCP host may require its own authentication configuration.
• Never paste bearer tokens or Service Keys into a public prompt, README, issue, generated source file, or committed environment file.
• Use test recipients and test assets before production sends.
• Confirm the environment, tenant, sender, recipient, and payload before every irreversible external action.
• Record message IDs, correlation IDs, request timestamps, and acceptance status for operational troubleshooting.
• Configure Contact Policy and opt-out handling for promotional messaging.

Troubleshooting

The skill is not discovered

Verify that the folder contains SKILL.md, that the folder name is lowercase and hyphenated, and that the skill was copied into the correct host or workspace skills directory. Restart or reload the AI tool after installation.

The plugin is installed but commands are missing

Check that the plugin directory includes the matching manifest, skills/, commands/, and rules/ paths. For a local plugin, reload the host after copying files and verify the plugin root rather than its parent directory.

Sandbox setup asks for MCP authentication

Stop and re-check the environment selection. Sandbox setup should not use the Production MCP server or its OAuth credentials. Use the documented Sandbox API workflow instead.

Sender discovery returns no compatible sender

Do not invent a sender ID. Ask the tenant administrator to approve or expose a sender for the intended use case, then repeat discovery.

A request is queued but no message is received

Queued means accepted for processing, not delivered. Check the recipient or device registration, sender configuration, channel credentials, provider status, webhook receipts, and tenant logs.

Source documentation

• Public Webex Skills repository
• Installable Webex Connect AI plugin
• Webex Connect skill catalog
• Plugin releases and changelog
• Webex Connect API overview: https://developers.webexconnect.io/reference/apioverview
• Webex Connect API authentication: https://developers.webexconnect.io/reference/api-authentication
• Webex Connect regional endpoints: https://developers.webexconnect.io/reference/know-your-api-endpoints
• Webex Connect Digital Channel API endpoints: https://developer.webex.com/webex-connect/docs/webex-connect-api-endpoints-and-regions
• Connect CPaaS MCP Server: https://developer.webex.com/mcp/docs/connect-mcp-server