Menu
Mattering Automations connector

A tawk.to MCP for the half of tawk.to that has an API

Connect one tawk.to REST API key and 11 nodes land on the canvas: list the properties the key can see, read chat history and tickets, read and create Knowledge Base articles and categories, subscribe and unsubscribe webhooks, and call any other method by name. Read this page before you plan around it, because the API is smaller and harder to get access to than the product is, and both of those facts are stated here rather than discovered later.

11 ready-made nodes One credential to connect Works with Claude, ChatGPT, and Matter

What the tawk.to MCP connector does

The tawk.to connector for Mattering Automations is 11 nodes on tawk.to REST API v1: 6 reads and 5 actions, of which one is a custom method node. Every call is a POST to a dot-named method with its parameters in a JSON body, which is worth knowing because a request carrying no body at all is rejected outright with a validation error naming the body itself. The connector always sends at least an empty object, including through the custom method node, so that trap is handled for you.

Every node speaks MCP, the open protocol for connecting AI assistants to tools, so Claude, ChatGPT, and any MCP-capable assistant can use them, alongside Matter, the AI assistant built into the platform. An agent can pull a week of tickets with tawk.to: List Tickets, summarise the themes, and draft the Knowledge Base article that would have prevented half of them with tawk.to: Create Knowledge Base Article, which defaults to a draft so nothing reaches your public help centre without a person saying so.

Two facts about access have to be separated, because merging them is how people end up disappointed. The tawk.to live chat product is genuinely free: creating an account, adding a property and running the widget on your site costs nothing. The REST API is a different thing and it is not self-serve. Its reference documentation is served behind a password, and a key can only be created once tawk.to approves an access request submitted through a form on its website. The public record on how long that takes is contradictory: tawk.to has said it typically takes under a day, and there are users on its own forum who applied and never heard back, one of them describing a wait of nearly a year. So plan for the key as something you request now and receive on a timeline you do not control.

The size of this connector is a consequence of that gate rather than a scoping decision, and it deserves the same honesty. Because the reference is not public, there is no published list of methods anywhere, and the ten that are publicly evidenced use three different naming conventions between them, so the rest cannot be guessed. Shipping a node that calls a route which does not exist is worse than shipping nothing there, so the typed nodes stop exactly at the evidence line. tawk.to: Custom API Method (any method) is what carries the rest: anyone holding docs access can call any method by name through it, with the Basic auth, the mandatory JSON body and the error handling already correct. The API is not ten methods wide; ten is what the public record proves.

Four things a reader will reasonably assume work do not exist on this API, and tawk.to has confirmed each of them on the record. There is no ticket creation. There is no contact create or sync, because visitor identity is a job for the JavaScript widget rather than the REST API. There is no way to fetch live or in-progress chats, so chat reads are history and real time comes from webhooks or not at all. And ticket replies are not returned, so a ticket read carries its opening message only. None of that is a gap in this connector; it is the shape of the API, said up front.

One connection, three surfaces

  • Visual automations. Drag tawk.to nodes onto the canvas and wire them to triggers, conditions, and other apps.
  • AI agents, yours or ours. These nodes speak MCP, the open protocol, so Claude, ChatGPT, and any MCP-capable assistant can use them, and Matter, the agent built into the platform, uses them natively.
  • The whole platform. Combine with native tasks, projects, and other connectors, no middleware in between.

Every tawk.to node, ready to drop on the canvas

11 nodes, generated from the connector itself so this list always matches what ships. Expand a category to browse.

Read and query nodes 6

tawk.to: Get Current Agent

Return the agent the REST API key belongs to. This is the vendor's own worked example and the cheapest way to prove a key works, so it is this connector's auth probe. It takes no parameters, but tawk.to still requires a JSON body, so the node always sends an empty object; a request with no body at all fails with validation_error on 'body'. A POST despite being a pure read: every tawk.to method is a POST.

tawk.to: Get Knowledge Base Article

Fetch one Knowledge Base article by id. Two quirks are worth knowing before you debug this node: tawk.to requires the category id here as well as the article id, and it rejects a real Knowledge Base site id, accepting only 'primary'. Both are the vendor's behaviour, reproduced faithfully. Asking for the content block has also been reported to return a server error while the meta block succeeds, so a 500 here is a known gap rather than a configuration mistake.

tawk.to: List Chats

List past chats for a property, optionally narrowed to a date range. This is chat HISTORY, not a live feed: tawk.to confirms in-progress chats are not returned, so use a webhook if you need real time. A call returns at most 50 chats and there is no cursor, so walk longer histories by moving the dates. Chat payloads carry visitor names, email addresses and full message text, so treat the output as personal data. tawk.to publishes no cursor and no page token: walk history by moving the start and end dates, not by paging.

tawk.to: List Knowledge Base Articles

List the Knowledge Base articles on a property, optionally narrowed to one category. Use it to find the article id that Get Knowledge Base Article needs, or to audit which articles are still drafts. Unlike the single-article read, this method accepts a real Knowledge Base site id as well as 'primary'.

tawk.to: List Properties

List every property the API key's account can see. A property is one chat widget on one site, and its id is required by every other tawk.to node, which makes this the root of the connector's picker chain. Because the key is account-scoped rather than property-scoped, this returns the whole estate. Takes no required parameters; the node sends an empty JSON body when none are set, which tawk.to requires.

tawk.to: List Tickets

List tickets for a property, filterable by status, date range and tags. Each ticket carries its opening message only: tawk.to confirms follow-up replies are not returned by this method and has no committed timeline for adding them. Ticket creation is not available on the REST API either, so this is a read-out, not a round trip. tawk.to publishes no cursor and no page token: walk history by moving the start and end dates, not by paging.

Action nodes 5

tawk.to: Create Knowledge Base Article

Create a Knowledge Base article on a property. The node assembles tawk.to's three-deep payload for you from flat fields, and deliberately omits the category NAME that the vendor's documentation still shows: sending it is rejected, and tawk.to has confirmed the docs are wrong there. Defaults to a draft, so a test run cannot publish to real readers - which matters because the REST API publishes no article DELETE, so removing an article means opening the dashboard. The slug must be unique across the Knowledge Base or the call fails with a server error rather than a helpful message.

tawk.to: Create Knowledge Base Category

Create a Knowledge Base category on a property, and return its id - which is worth having, because tawk.to publishes no category lister, so this node and the dashboard URL are the two ways to learn a category id. The node assembles the nested payload from flat fields. There is no category delete on the REST API, so cleaning up a test category means using the dashboard.

tawk.to: Create Webhook

Subscribe a URL to tawk.to events on a property, and return the hook id. This is how you get real time out of tawk.to: chats cannot be polled live, so a webhook is the only way to react to one while it is happening. tawk.to signs every delivery with HMAC-SHA1 in the X-Tawk-Signature header using the property's webhook secret, retries a failed delivery for up to 12 hours, and treats no response within 30 seconds as a failure. Keep the returned hook id: Delete Webhook needs it and there is no lister to recover it from.

tawk.to: Custom API Method (any method)

Call any tawk.to REST method by name. tawk.to publishes its REST reference only to accounts approved through a manual beta form, so the typed nodes here cover the ten methods evidenced in public sources and this node covers everything else for anyone holding that access. Every tawk.to method is a POST and requires a JSON body, so send at least an empty object.

tawk.to: Delete Webhook

DESTRUCTIVE. Remove a webhook subscription from a property. There is no restore and no undo: the hook id is gone, a replacement gets a new one, and anything keyed to the old id breaks. tawk.to also publishes no webhook LISTER on the REST API, so a hook id you did not record when you created it can only be found in the dashboard. Deliveries stop immediately, which means events that happen after this call are lost rather than queued.

Connecting tawk.to takes one credential

The key is created in the tawk.to dashboard under your profile image, then Edit Profile, then REST API Keys, then Create Key. If that tab is not there, your account has not been granted REST API beta access yet, and its absence is exactly what people report as not being able to find the screen. Access is requested through the REST API beta access form on tawk.to and granted manually by tawk.to; approval also brings the username, password and link for the reference documentation, which is served behind a password rather than published. One detail changes how every node is configured: the key belongs to the ACCOUNT and not to a property. One key sees every property the account can reach, which is why the property is chosen on each node from a picker rather than stored on the credential, and why the credential itself holds exactly one field. There is no password to enter, and typing one produces a 401 on every method. tawk.to publishes no rate limit, no requests-per-second ceiling and no retry contract, so this page states no number; its own guidance is simply to plan for retries with backoff.

1

Request REST API access from tawk.to using its REST API beta access request form, under the identity you intend to run automations as. Approval is manual and is not immediate.

2

Once approved, open the tawk.to dashboard, click your profile image at the lower left, choose Edit Profile, open REST API Keys, and click Create Key. Copy the key.

3

In Mattering Automations, open Connectors, click Add Connector, choose tawk.to, and create a credential with that key. There is only one field and no password.

4

Click Test Connection. Mattering posts to the agent endpoint with an empty body, which is tawk.to's own worked example and the only method that needs no property. A pass proves the key is valid and the Basic construction is right; it does not prove any particular property is reachable or that the Knowledge Base methods will work, since those have server-side faults of their own.

5

All 11 tawk.to nodes appear in the canvas palette. Run tawk.to: List Properties first: it returns every property the key can see, and it is the source for the property picker on the other eight nodes that need one.

Common uses for the tawk.to MCP

Get chat history out of tawk.to and somewhere you can use it

tawk.to: List Chats reads past chats for a property, optionally narrowed to a date range, and that is the whole of the bulk export story. A scheduled workflow can walk last week in date windows, write the rows into a Mattering database, and summarise the themes into a Mattering Comms thread so the team reads a digest rather than a dashboard. Two constraints shape the design rather than block it: a call returns at most 50 chats and there is no cursor, so a longer history is walked by moving the dates and stopping when a window comes back short.

Ticket reporting without a ticket API you do not have

tawk.to: List Tickets reads tickets for a property filtered by status, date range and tags, which is enough for a weekly backlog report, an ageing check, or an alert when open tickets carrying a given tag cross a threshold. Be clear about the boundary before building on it: this is a read-out and not a round trip. tawk.to publishes no ticket creation on the REST API and confirms follow-up replies are not returned, so a ticket row carries its opening message and its metadata. Reporting, routing and alerting all work; closing the loop back into tawk.to does not.

Build the Knowledge Base from the questions people actually ask

tawk.to: List Knowledge Base Articles and tawk.to: Get Knowledge Base Article read what is already published, and tawk.to: Create Knowledge Base Article writes new ones, assembling tawk.to's three-deep payload from flat fields. It defaults to a draft, which matters more here than usual: the REST API publishes no article delete, so an article created by mistake has to be removed by hand in the dashboard. tawk.to: Create Knowledge Base Category returns a category id, which is worth capturing, because tawk.to publishes no category lister either and that node and the dashboard are the only two ways to learn one.

Wire real-time events in, since they cannot be polled

tawk.to confirms in-progress chats are not available on the REST API, so webhooks are the only real-time route. tawk.to: Create Webhook subscribes a URL to a property's events and returns the hook id; point it at a Mattering inbound webhook trigger and a workflow runs the moment a chat starts, a chat ends, a transcript is created or a ticket is raised. tawk.to signs every delivery with HMAC-SHA1 in a signature header, retries a failed delivery for up to 12 hours and treats no response within 30 seconds as a failure. Keep the returned hook id somewhere durable: tawk.to: Delete Webhook needs it and there is no lister to recover it from.

Know what the account actually contains

tawk.to: List Properties returns every property the key can see, which on an account that has accumulated sites over the years is frequently more than the person running the automation expects. It is the root of the connector's picker chain, so every other node's property field is filled from it rather than typed, and it doubles as an inventory: write the list into a Mattering database on a schedule and a new property appearing is something a workflow can notice. tawk.to: Get Current Agent tells you which agent the key belongs to, which is the fastest way to confirm you are holding the key you think you are.

Reach the methods the public record does not name

tawk.to: Custom API Method (any method) calls any tawk.to method by name with a JSON body you supply. On most connectors a catch-all is a footnote; here it is the second most important node, because the reference is password-protected and the typed nodes cover the ten methods public sources evidence rather than everything the API has. tawk.to advertises capabilities whose method names are nowhere public, including creating properties, inviting team members, listing property members, chat statistics by agent and tag, and the list and update halves of webhook management. If your approved documentation names one of those, this node calls it with the Basic auth, the mandatory body and the error handling already correct.

Example: A Monday morning support digest built from a date window

Built on the Automations canvas in a few minutes, or described in plain language to Matter and built for you.

1

Trigger: a weekly schedule fires on Monday morning.

2

tawk.to: List Properties runs once, so the workflow iterates every property the key can see rather than a hard-coded id.

3

tawk.to: List Tickets reads each property with the start and end dates set to the previous week, filtered to the statuses you care about. There is no cursor here, so the window IS the pagination.

4

tawk.to: List Chats reads the same window for the same property, remembering that a call returns at most 50 chats, so a busy property is walked in shorter windows rather than one long one.

5

A transform step selects only the fields the digest needs: counts, tags, statuses and timestamps. Chat and ticket payloads carry visitor names, email addresses and full transcripts, so forwarding whole objects into a summary or an AI step means forwarding personal data with them.

6

An AI step groups the week into themes and drafts the one Knowledge Base article that would have prevented the most repeats.

7

tawk.to: Create Knowledge Base Article writes that draft to the property's Knowledge Base with its status left at draft, and a Mattering task opens for a person to edit and publish it. Publishing by hand is the right call here anyway, since the REST API cannot delete an article afterwards.

tawk.to MCP questions, answered

What tawk.to operations are supported?

11 nodes on tawk.to REST API v1: 6 reads and 5 actions. Reads are the current agent, properties, chat history, tickets, a Knowledge Base article list and a single Knowledge Base article. Actions are create a Knowledge Base article, create a Knowledge Base category, create a webhook subscription, delete a webhook subscription, and a custom method node that calls any tawk.to method by name. That is ten typed methods plus the catch-all, and it is the ceiling of the public record rather than a scoping choice: tawk.to serves its REST reference behind a password, so no complete method list is published anywhere, and the ten known methods use three different naming conventions, which makes the rest unguessable. Anything your approved documentation names is reachable through the custom method node today.

Does this work with Claude, ChatGPT, or other AI assistants?

Yes. Mattering exposes these nodes over MCP, an open protocol for connecting AI assistants to tools, so any MCP-capable assistant, including Claude and ChatGPT, can connect to your workspace and use them. Matter, the assistant built into the platform, uses the same nodes natively. One caution specific to this connector: chat and ticket payloads carry visitor names, email addresses and full transcripts, so select the fields a step actually needs rather than handing a whole chat object to an assistant's context.

Do I need special access to use the tawk.to REST API?

Yes, and this is the single most important thing to know before planning around it. The tawk.to product itself is free and needs no approval. The REST API is separate and is not self-serve: its reference documentation is password-protected, and the REST API Keys tab does not appear in your profile at all until tawk.to grants access in response to a request form on its site. Approval is manual and the public record on timing is contradictory, ranging from tawk.to saying it typically takes under a day to users on its own forum reporting that they applied and never heard back, one of them after nearly a year. Submit the request early and treat the key as arriving on a timeline you do not control.

Can an agent reply to a chat, create a ticket, or see chats as they happen?

No to all three, and tawk.to has confirmed each on the record rather than us inferring it. Ticket creation is not available on the REST API and is described by tawk.to as a future consideration with no timeline. Live or in-progress chats cannot be fetched; the chat method lists a property's chat history only, so real time comes from a webhook subscription or not at all. Ticket replies are not returned, so a ticket read carries its opening message only. And there is no contact create or sync endpoint, because identifying a visitor is a job for tawk.to's JavaScript widget rather than its REST API. What this connector does well is read history out, report on it, and build Knowledge Base content; what it cannot do is drive a conversation.

Does tawk.to publish an official MCP server?

No. We re-checked on 20 August 2026 rather than assuming. tawk.to's developer portal covers its JavaScript API, webhooks and Zapier and mentions MCP nowhere, its help centre returns nothing for it, and its npm scope publishes widget plugins rather than an MCP server. The clearest single data point is on tawk.to's own forum, where a user opened a thread asking explicitly for an API or an MCP to reach chat history and staff answered with the REST API beta form and the JavaScript API without mentioning an MCP server at all. So there is no first-party server here to defer to. What this connector adds beyond an assistant tool is that the same methods run as scheduled workflow steps with nobody present, chain with every other connector in one workflow, and hold the credential once in a shared workspace store rather than in one person's local config.

Is the API key per property or per account?

Per account, which surprises people who expect a per-site key. The key is minted on a profile screen rather than a property screen, and it can see every property the account reaches, which on an older account is often more than expected. That is why the credential holds exactly one field and the property id is a per-node field with a picker sourced from tawk.to: List Properties. One credential therefore covers your whole estate, and choosing the wrong property is a node-level mistake rather than a credential-level one, so check the picker value on any node that writes.

How do I page through a long chat or ticket history?

By moving the dates, because there is no pagination in this API at all. tawk.to publishes no cursor, no page token and no offset, and no response says whether more rows exist. A chat list returns at most 50 results per call, confirmed by a user who extracted history with documentation access. The total on a list response is a count of matched records rather than a page count. So the pattern is to walk start and end dates in windows small enough that a window never comes back at the cap, and to keep the window arithmetic in the automation where it is visible. tawk.to also publishes no rate limit, no 429 contract and no backoff window, so this page quotes no number; treat throughput as unknown rather than generous and pace bulk reads.

What personal data comes back from these nodes?

A lot, and it is worth designing around rather than discovering in a run log. A chat payload carries the visitor's name, email address, city and country, plus the complete verbatim transcript including links to any files they uploaded. A ticket payload carries the requester's name and email and the full body of their message. Every sample output in this connector's palette is synthetic for that reason. The practical rule for a workflow is to select the fields you need rather than forwarding a whole chat or ticket object into a chat message, a spreadsheet, a summary or an AI step, and to be deliberate about which downstream systems a transcript is allowed to reach.

11 nodes waiting on the canvas

Put tawk.to to work inside your workspace

Connect once, then use every node in visual automations and AI agent workflows across the Mattering platform.

All product names, logos, and brands are property of their respective owners and are used for identification purposes only. Use of these names, logos, and brands does not imply endorsement or affiliation. tawk.to is a product of its respective owner; Mattering is an independent platform.

Start Free Explore our Products