v268 MCP & Integrations as LLM Tools 🛠️
"I may be synthetic, but I'm not stupid." - Bishop, Aliens 🤖
SentiOne Automate version 268 was released on 19.11.2026. This release lets the LLM Say block call MCP servers and your existing integrations as tools on its own, turns PDFs, images and web pages into Knowledge Base documents in one step, and keeps WebChat conversations in sync across browser tabs. It also introduces the new generation of OpenAI models ahead of the shutdown of their predecessors.
MCP and Integrations as Tools in LLM Say
The LLM Say block can now use real tools during a conversation without any extra wiring in the flow. When the model decides a tool is needed, Automate calls it, passes the result back to the model, and the conversation simply continues.
- MCP integrations. MCP servers are configured once as reusable integrations, in a dedicated tab of the Integrations module. While setting one up you can fetch the tools the server offers and choose exactly which of them the bot may use. In the LLM Say block you then just pick the configured MCP integrations from a list.
- HTTP integrations as tools. Existing integration requests can be exposed to the model as tools: their input parameters become the tool's inputs, and their output values are returned to the model.
- Automatic execution. No more loops of Integration and Custom blocks to run tools and feed results back. The block handles the whole round trip for you.
Read more about LLM Say and Integrations.
Documents from PDFs, Images and Web Pages
Creating a Knowledge Base document no longer requires preparing a spreadsheet first. The document form now lets you choose how to import content:
- Unstructured files. Upload several files at once, including PDFs, images, XLSX and CSV, and they are converted into a searchable document automatically.
- URLs. Add one or more web pages and their content is imported in the same way.
- Structured XLSX. The existing spreadsheet import stays available, including for environments without access to cloud AI services.
Conversion runs in the background with the usual indexing progress, and the list shows which files a document was built from. Updating an existing document still uses the structured XLSX format. Read more about Documents.
WebChat Across Multiple Tabs
When a user has the chat open in several browser tabs, every tab now shows the same conversation. Messages sent in one tab, and the bot's replies to them, appear in all the others, and starting or ending a conversation in one tab is reflected everywhere. If two tabs send a message at the same moment, the user sees a friendly hint instead of an error.
New OpenAI Models
The gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna models are now available in LLM and LLM Say blocks, with and without tools, and in LLM-based NLU. OpenAI shuts down gpt-5, gpt-5-mini, gpt-5-nano and o3 on 11.12.2026, so these models are no longer offered. Blocks that still use one of them show it as no longer available. Recommended replacements:
- gpt-5 and o3 → gpt-5.6-sol
- gpt-5-mini → gpt-5.6-terra
- gpt-5-nano → gpt-5.6-luna
Please switch affected blocks before the shutdown date. See the list of supported models.
💊 IMPROVEMENTS
- [LLM] LLM Say can call MCP servers and HTTP integrations as tools, with results returned to the model automatically
- [Integration] MCP servers configured as reusable integrations, with a choice of which tools to expose
- [Knowledge Base] Documents created from multiple PDFs, images, spreadsheets or URLs in one step, with source files shown on the list
- [WebChat] Conversation kept in sync across multiple browser tabs
- [LLM] gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna supported; gpt-5, gpt-5-mini, gpt-5-nano and o3 removed ahead of their 11.12.2026 shutdown
- [Chat Tester] Transfers to a route can be tested by choosing the Live or Candidate route variant in the tester menu
- [Flow] A route's Transfer to Flow block can be published when the selected flow exists in at least one of the route's projects
- [NLU] Polish date and time phrases using "godz." or a bare hour after "po", "przed" or "od" now resolve to the full date and time; phrases like "po 15" return an interval with a lower bound
- [Flow] Blocks that wait for user input are marked on the canvas
- [Flow] Smoother zooming on the flow canvas
- [Security] Updated third-party components to address known vulnerabilities
🐛 FIXES
-
[LLM] Reasoning models from OpenAI-compatible providers no longer fail when they return a response without text content
-
[LLM] The waiting message and the model's answer in LLM Say stay separate messages; mobile apps no longer show the following bot message twice
-
[LLM] Closing the LLM Integration block warns about unsaved changes only when something was changed
-
[NLU] More reliable intent detection in very small NLUs with only a few intents or phrases
-
[Knowledge Base] Users with read-only Knowledge Base access no longer see create, edit and delete actions for documents
-
[Chat Tester] Long user messages are no longer displayed without a left margin
-
[UI] The copy Project ID icon on pinned projects stays in place after clicking
Follow the changelog for more exciting new features! 👋