1. All Collections >
  2. Integrations >
  3. Model Context Protocol (MCP)

Model Context Protocol (MCP)

Avatar
Joshua Lim
4 min read

The respond.io MCP server lets AI assistants, such as Claude or ChatGPT, act on your Workspace using natural language. The hosted MCP server runs on respond.io's own infrastructure. It authorizes access with OAuth, so you don't install anything, host anything, or share a credential.

Before you begin

  • You need a respond.io account with access to the Workspace you want to connect.

  • The connection only exposes what your respond.io user role already allows. Connecting an AI assistant does not grant it any additional access.

Connect with a native connector

  1. In your AI client, open its connector or plugin directory and select respond.io.

  2. Sign in with your respond.io account when prompted.

  3. On the Connect with respond.io MCP screen, select the Workspace to authorize. If you only have access to one Workspace, it's selected automatically.

  4. Review the permissions listed, then click Allow.

    Connect with respond.io MCP consent screen with a Workspace selected and the Allow button active
  5. Once authorized, check whether the connection was successful in your AI client.

Connect with the MCP server URL

Some AI clients don't yet list respond.io as a native connector. Add the connection manually instead:

  1. In your AI client, open the option to add a custom MCP server.

  2. Enter the respond.io MCP server URL: https://mcp.respond.io/mcp

  3. Sign in and authorize the connection the same way as the native connector flow above.

What the hosted MCP server supports

Once connected, your AI assistant can work with:

Category

What you can do

Contacts

- Look up a specific Contact by ID, email, or phone
- Search and filter across all Contacts by criteria
- Create, update, or delete a Contact
- Manage tags and Lifecycle stages
- View the Channels connected to a Contact

Messaging

- Send messages, templates, and attachments
- Review message history and delivery status

Conversations

- Assign or unassign
- Open or close
- Add internal comments
- View the closing note categories used when closing a conversation

AI Agents

- List, create, update, activate, deactivate, and delete AI Agents (deleting reassigns the agent's Contacts)
- Enhance an AI Agent's prompt
- Configure an AI Agent's in-conversation actions
- List knowledge sources and add them to an AI Agent

Workspace settings

- View team members, teams, Channels, and Lifecycle stages
- View and create custom fields
- View, create, update, and delete tags

Workflows

- View Workflows, including their status and settings   
- Create a Workflow from scratch or from a respond.io template
- Inspect and edit a Workflow
- Validate, visualize, and test a Workflow
- Publish, stop, or delete a Workflow
- Trigger a published Workflow that uses a Shortcut trigger

Your account

View your profile and the Workspace and Organization you're connected to

Every action is scoped to what you're already allowed to do in respond.io.

Building Workflows through MCP

Ask your AI assistant to build or change a Workflow in plain language.

Note: Workflows are available on the Growth plan and above. HTTP request steps require the Advanced plan or above.

Create a Workflow

  • Describe the automation you need, or start from a respond.io template. The Workflow starts as a draft.

  • Limits: 1 trigger, up to 100 steps, up to 9 branches plus Else, and up to 10 conditions per branch.

  • Scheduled and time-based triggers aren't supported.

  • Example: "Create a Workflow that adds a tag to new Contacts and assigns them to the Sales team."

  • Your AI assistant can also edit Workflow Settings, such as exit conditions and allowing users to stop the Workflow for a Contact.

Validate and test a Workflow

  • Validation flags missing or invalid settings, invalid references, unreachable steps, and dead-end branches.

  • Tests run on a Channel you choose.

Important: Tests can have real side effects, such as messages sent to real Contacts.

  • Example: "Check my welcome Workflow for errors, then test it on my WhatsApp Channel."

Publish, stop, or delete a Workflow

  • Your AI assistant asks for your confirmation before publishing, stopping, or deleting a Workflow.

Important: Editing a published Workflow stops it first. Contacts in it leave and aren't re-enrolled when you republish.

  • Example: "Publish my welcome Workflow."

Trigger a published Workflow

  • Your AI assistant can trigger a published Workflow for a Contact when the Workflow's trigger is Shortcut.

  • If the Shortcut has input fields, your AI assistant asks you for those values first.

  • Example: "Trigger my order follow-up Workflow for Sarah."

Managing AI Agents through MCP

Two of these capabilities change an AI Agent's setup rather than a live conversation. Ask your AI assistant for either one in plain language.

Enhance an AI Agent's prompt

This returns an AI-improved version of an existing prompt. The result is a suggestion โ€” your AI Agent keeps its current prompt until you apply the change.

Example: "Enhance the instruction section of this prompt."

Configure in-conversation actions

In-conversation actions are the platform actions an AI Agent takes while it replies to a Contact. Configuring them sets which actions the AI Agent is allowed to use.

Example: "Set up my support agent to close conversations, assign to Sales, update the Lifecycle stage and Contact fields, add tags, trigger my follow-up Workflow, and add comments."

For the full list and how each action behaves, see Using AI Skills.

Managing your connection

  • Each connection is authorized for one Workspace. To connect a different Workspace, reconnect and select it during sign-in.

  • Permissions update immediately if your respond.io role changes or your Workspace access is removed. No separate token exists to expire or revoke.

FAQ and troubleshooting

Can I still use the local, self-hosted MCP server?

Yes. The self-hosted setup is unchanged and continues to work with a Developer API token. See the respond.io MCP developer documentation for those steps, or Developer API for how to generate a token.

Can I connect more than one Workspace at once?

No. Each connection covers one Workspace โ€” reconnect to switch.

Does connecting an AI assistant give it more access than I have?

No. Every action is limited to your existing respond.io permissions.

Can I use these actions without connecting an external AI assistant?

Yes. Copilot, the in-app assistant in respond.io, can use the actions in the table above. Copilot asks for your approval before making any change. See Using Copilot.

Does my AI assistant publish a Workflow without asking?

No. Your AI assistant asks for your confirmation before it publishes, stops, or deletes a Workflow. Each of these actions needs its own confirmation.

What happens to Contacts when my AI assistant edits a published Workflow?

Your AI assistant stops the Workflow before editing it and asks you to confirm first. Contacts in the Workflow leave when it stops. Republishing doesn't re-enroll them.

Can my AI assistant add a Google Sheets step?

Your AI assistant adds the step and flags whatโ€™s missing. Connect Google and pick the sheet in the Workflow Builder to finish it. The same applies to file attachments.

Does enhancing a prompt update my AI Agent?

No. The action returns a suggestion only. Your AI Agent keeps its current prompt until you apply the change.

Do MCP actions consume AI Credits?

No. The actions listed in the table above don't consume AI Credits. AI Agents you create or update through MCP still consume AI Credits when they reply to Contacts. See AI Credits for details.

What happens if the connection fails?

You'll see a screen stating the connection wasn't established, with a reason for the failure. Close the tab and try connecting again.

What's next?

Share this article
Telegram
Facebook
Linkedin
Twitter

Can't find what you're looking for? ๐Ÿ”Ž