Beta feature: The Olark MCP server is currently in beta. Features may change, and AI-generated results can occasionally be incomplete or incorrect. Please review outputs carefully—especially before approving changes to your bots or Olark settings—and use the feature at your own discretion.

We’d love to see what you’re doing with it. Schedule a short call with our team to share your workflow, feedback, or ideas.

The Olark MCP server lets compatible AI assistants securely work with your Olark data and tools. Once connected, you can use natural-language prompts to analyze chat transcripts, identify trends, and manage Olark AI bots.

For example, you could ask:

What questions have customers asked most often this week?

Or:

Review my draft bot and suggest changes that would help it answer our pricing questions more accurately.

What is an MCP server?

MCP stands for Model Context Protocol. It is an open standard that allows AI applications to connect to external tools and data sources.

Without an MCP connection, an AI assistant only knows what you type or upload. With the Olark MCP server connected, the assistant can securely request information from Olark or perform supported Olark actions on your behalf.

You can think of MCP as a universal connection between an AI assistant and another application:

  1. You ask the AI assistant a question or give it a task.
  2. The assistant decides which Olark tool it needs.
  3. Olark checks your identity and permissions.
  4. The tool returns the relevant information or performs the requested action.
  5. The assistant explains the result in plain language.

The AI assistant does not receive unrestricted access to Olark. It can only use the tools exposed by the server and the data your Olark account is authorized to access.

Learn more in the introduction to Model Context Protocol.

Which AI applications can connect to it?

The Olark MCP server is a remote MCP server. It may be connected to AI applications that support remote MCP servers, Streamable HTTP, and OAuth authentication.

Examples include:

Support may depend on your AI provider, subscription, workspace settings, and administrator policies. Some applications support reading data but not actions that change data.

Note: Microsoft Copilot Studio is not the same as every consumer or Microsoft 365 Copilot experience. Check the documentation for the particular Microsoft product you use.

Before you connect

You will need:

  • An Olark account with access to the site you want to use
  • Permission to view transcripts or manage bots, depending on what you want the assistant to do
  • A compatible AI application
  • Permission to add a custom MCP server or connector in that application

Your organization’s AI workspace administrator may need to approve the connection.

How to connect

The labels differ between AI applications, but the general process is the same.

  1. Open your AI application’s settings.

  2. Find MCP servers , Connectors , Apps , or Tools.

  3. Choose the option to add a custom or remote MCP server.

  4. Enter the following server URL:

    https://mcp.olark.com/mcp

  5. If you are asked for an authentication method, select OAuth.

  6. Follow the sign-in and authorization prompts.

  7. Review the Olark tools the application discovers.

  8. Enable the connection for your conversation, project, or agent.

Because each AI application presents connections differently, see the separate setup guide for your application for exact instructions.

What can the Olark MCP server do?

The tools available to you depend on your Olark permissions and which features are enabled for your account.

Analyze chat transcripts

The assistant can search a filtered group of Olark transcripts and analyze their conversation content.

You can filter transcripts by:

  • Words or phrases used in the conversation
  • Visitor name
  • Visitor email address
  • Tags
  • Start and end dates

Transcript results can include conversation text, transcript links, timestamps, tags, transcript type, and the visitor’s first-touch URL when available.

Common uses include:

  • Finding frequently asked questions
  • Identifying recurring complaints
  • Reviewing sales or buying signals
  • Detecting unresolved conversations
  • Assessing agent performance
  • Finding gaps in website or bot content
  • Understanding which landing pages lead to conversations

Each search returns up to 100 matching transcripts, though most AI models are smart enough to request the next 100 transcripts in the range.

Review your bots

Where bot-management tools are enabled, the assistant can:

  • List your published and draft bots
  • Read a bot’s current behavior and settings
  • Review its company name and tone instructions
  • List the knowledge sources available to it
  • Explain how the bot is currently configured

This is useful for auditing a bot before deciding whether to change it.

Create and update draft bots

The assistant can create a new draft bot:

  • From your website domain
  • By copying an existing bot
  • From behavior and settings you provide

It can also update an unpublished draft’s:

  • Name
  • Company name
  • Tone and voice
  • Conversation behavior
  • Contact-capture fields
  • Proactive engagement rules
  • Other supported bot settings

Published bots cannot be edited directly. To change a published bot, create a new draft copy.

For safety, ask the assistant to explain its proposed changes before applying them.

Manage bot knowledge sources

The assistant can list, add, or remove knowledge sources for a draft bot.

Supported sources include:

  • A website domain that the bot can search
  • Markdown content, such as a product guide, policy, or FAQ

Adding a source does not guarantee that every answer will be correct. Test the draft after changing its knowledge.

Preview and publish bots

The assistant can generate a preview URL so you can test a bot in your browser.

It can also publish a draft after you explicitly confirm the action. Publishing makes the bot available in the Olark application, but it does not necessarily make the bot live on your website. You may still need to open the bot in Olark and enable or deploy it to your widget.

Example prompts

You do not need to know the names of the underlying MCP tools. Describe the result you want in ordinary language.

  • “What are the most common topics in chats from the last 30 days? Group them into sales, support, product feedback, and other.”
  • “How many chats this week mentioned pricing, refunds, or a specific product question?”
  • “What are customers most confused about on our website, based on chats from the last two weeks?”
  • “Summarize the recurring complaints from the last two weeks so I can share them with the product team.”
  • “Which pages generated the most chats last month? Use the first-touch page when it is available.”
  • “Find conversations that ended with a frustrated or unresolved customer and explain what appears to have gone wrong.”

Find sales opportunities

  • “Find chats from the last 60 days that show strong buying intent. Include the visitor, reason you classified it as buying intent, and a link to the transcript.”
  • “Which chat contacts from the last month appear most likely to become leads?”
  • “Find chats mentioning pricing, demos, implementation timelines, or contract questions.”
  • “Which conversations suggest an urgent purchase need or a short decision timeline?”

Review support and agent performance

  • “Based on recent transcripts, what are our human agents doing well and where could they improve?”
  • “Which agents appear to have the strongest response quality across the chats available to you?”
  • “Find possible off-brand, inaccurate, or compliance-sensitive responses in chats from the last month.”
  • “Turn 10 representative support chats into anonymized training scenarios for a new support agent.”

The transcript tools provide conversation content and timestamps, but they may not provide every operational metric used by your reporting system. Ask the assistant to distinguish measured data from its qualitative assessment.

Review a bot

  • “List my bots and tell me which are drafts and which are published.”
  • “Review my current draft bot. Summarize its tone, contact-capture behavior, proactive engagement rules, and knowledge sources.”
  • “Does my draft bot have enough information to answer questions about pricing and refunds? Do not change anything yet.”
  • “Compare these two bots and explain the important differences in plain language.”

Create or improve a draft bot

  • “Create a draft bot based on example.com. Show me what you created, but do not publish it.”
  • “Suggest changes that would make my draft bot sound warmer and more concise. Let me approve the changes before you apply them.”
  • “Update my draft bot to ask for an email address when someone requests a demo. Explain the proposed behavior before changing it.”
  • “Create a new draft copy of my published bot and adapt it for sales qualification. Do not publish it.”

Manage bot knowledge

  • “List the knowledge sources available to my draft bot.”
  • “Add docs.example.com as a knowledge source for my draft bot.”
  • “Turn the following return policy into a Markdown knowledge source and add it to my draft bot.”
  • “Based on questions from the last month’s chats, recommend additions to the bot’s knowledge. Do not make any changes yet.”

Preview and publish

  • “Create a preview link for my draft bot using https://example.com as the website.”
  • “Give me a test plan for this draft bot, then provide its preview link.”
  • “Publish the draft bot named ‘Sales Assistant.’ Before publishing, tell me exactly which bot you will publish and ask me to confirm.”

Using Olark with a CRM or another connected application

The Olark MCP server only provides access to supported Olark data and actions. It does not automatically have access to your CRM, ecommerce platform, calendar, help desk, or analytics system.

Some AI applications can use more than one connector in the same conversation. If you separately connect your CRM or another business application, you may be able to combine its data with Olark transcripts.

For example:

  • “Find chats from the last 60 days that show buying intent but do not have a corresponding deal in our CRM.”
  • “Which contacts in our sales pipeline have chatted with us in the past three months?”
  • “Which chats are associated with our highest-value opportunities that are closest to closing?”
  • “Show which of my last 100 chats led to a purchase and the revenue associated with each.”
  • “Compare conversion rates for customers who chatted with us and customers who did not.”

These prompts require both the Olark connection and an appropriate CRM, ecommerce, or analytics connection. Whether the applications can be used together depends on your AI platform.

Tips for better results

Specify a time period

Instead of: “What are customers asking about?”

Try: “What were the five most common customer questions in chats between July 1 and July 15?”

Explain how you want the result organized

For example: “Group the results by topic. For each topic, include the number of matching chats, two representative examples, and links to the relevant transcripts.”

Separate facts from interpretation

For example: “Identify chats that may show frustration. Explain the evidence for each classification and label uncertain cases.”

Sentiment, intent, and response-quality assessments are AI interpretations unless they come from an explicit Olark field.

Ask for a proposal before making changes

For example: “Review the draft and propose changes. Do not update or publish anything until I approve the proposal.”

For example: “Analyze each week in June separately, then summarize the trends across the month.”

Security and privacy

Connecting an MCP server allows the selected AI application to request data from Olark within your permissions.

Before connecting:

  • Use only an AI application approved by your organization.
  • Review the application’s data-retention and model-training policies.
  • Confirm which Olark tools the application will be allowed to use.
  • Disable write actions if your AI application provides that option and you only need transcript analysis.
  • Avoid placing sensitive personal information directly in prompts unless necessary.
  • Review important changes before approving them.
  • Remove connections you no longer use.

Instructions inside transcripts, websites, or uploaded knowledge should be treated as untrusted content. Only connect the official Olark endpoint and use an AI application with appropriate safeguards.

Important limitations

  • Transcript searches return no more than 100 matching transcripts per request.
  • Transcript searching is limited to 60 requests per hour for each Olark site.
  • Results are limited by your Olark account permissions.
  • Some accounts may have transcript tools without bot-management tools.
  • AI-generated summaries and classifications can be incorrect and should be reviewed.
  • Revenue, deal stage, customer status, and other CRM information require a separate connected data source.
  • Published bots cannot be edited directly through MCP.
  • Publishing a bot does not automatically make it live on your website.

Troubleshooting

The Olark tools do not appear

Confirm that:

  • You entered the full server URL, including /mcp.
  • Your AI application supports remote MCP servers and Streamable HTTP.
  • The connection is enabled for the current conversation or agent.
  • Your workspace administrator has approved custom MCP connections.
  • You completed the Olark authorization process.

Try disconnecting and reconnecting the server if the application has cached an older list of tools.

I can search transcripts but cannot manage bots

Bot-management tools may not be enabled for your Olark account, or your Olark user may not have the required permission.

A CRM-related prompt does not work

Connect the CRM separately and make sure your AI application supports using multiple connectors in one task. The Olark MCP server does not provide CRM records by itself.

The bot was published but is not visible on my website

Publishing makes the bot available in Olark but may not enable it on your website. Open the bot in Olark and complete the deployment or widget configuration there.

Frequently asked questions

Does the MCP server give the AI access to my entire Olark account?

No. Access is limited by your authenticated Olark account, site, permissions, and the tools enabled for you.

Does connecting Olark mean my data will be used to train an AI model?

MCP itself does not determine how an AI provider stores or uses data. Review the privacy, retention, and model-training settings of the AI application you connect.