AI nodes: the tools you give to an agent
Ce contenu n’est pas encore disponible dans votre langue.
Generated on 2026-10-20 from a running server by tools/reference.mjs. Do not edit this page by hand.
The tools an agent can call. Each one is a resource that you plug into the place for tools of an AI Agent.
This page lists 16 nodes. How to read a node reference.
AI > Tools
Section titled “AI > Tools”AI-callable tools and tool gateways.
AI Agent HTTP Request Tool
Section titled “AI Agent HTTP Request Tool”Resource · in the catalogue: AI > Tools · type ai.tools.http.request
Lets an agent call a web API or read a web page: it reads by default (GET), and may also send (POST, PUT, PATCH, DELETE) when that is allowed on the node. A page is given as plain text, JSON as data.
When to use it. For an agent that works with a service that has a web API (looking up an order, creating a ticket), or that opens a web address it already has. Allow writing and limit the hosts to what the job needs. Not for finding pages: use a Web Search tool instead.
Effects. reads outside, changes something outside.
Fields
| Name | Type | Required | Default | What it is |
|---|---|---|---|---|
credentials | Credentials | no | The HTTP API connection whose key signs every call, when the API needs one. Connection of type http_api. | |
allow_write | Boolean | no | false | Let the agent send data (POST, PUT, PATCH, DELETE). Off: it only reads (GET, HEAD). |
allowed_hosts | Text | no | The hosts the agent may call, separated by commas (api.shop.com, docs.shop.com). Empty = any public address. | |
tool_name_prefix | Text | no | http | Prefix of the tool’s name, e.g. “shop” → shop_request. Needed when an agent has more than one HTTP Request tool connected: tool names must be unique per agent. |
max_chars | Number | no | 12000 | The most characters of an answer handed to the agent. |
rate_limit_profile | Choice | no | medium | How long to wait and try again when the service is busy (readings only). Choices: none, fast, medium, slow. |
Plugs into a place that accepts ai.node.tool.*.
AI Agent JavaScript Tool
Section titled “AI Agent JavaScript Tool”Resource · in the catalogue: AI > Tools · type ai.tools.code_javascript
Lets an AI agent run a JavaScript snippet (real, sandboxed Node.js) and use the result.
When to use it. To let an agent write and run a small piece of JavaScript for exact calculations or reshaping data (sorting, filtering, converting a list). The code cannot reach files or the internet: to read a web page, use AI Agent Web Scraper Tool (Direct) instead.
Fields
| Name | Type | Required | Default | What it is |
|---|---|---|---|---|
timeout_seconds | Number | no | 5 | Execution timeout per call, in seconds. |
Plugs into a place that accepts ai.node.tool.*.
MCP Server
Section titled “MCP Server”Resource · in the catalogue: AI > Tools · type ai.tools.mcp.server
Connects AI agents to a remote MCP server over HTTP.
When to use it. To give an agent the tools of an outside service that offers an MCP server: every tool that server lists becomes available to the agent. You need the server’s address (URL), and a token when the service asks for one.
Effects. reads outside, changes something outside.
Fields
| Name | Type | Required | Default | What it is |
|---|---|---|---|---|
server_url | Text | yes | Remote MCP HTTP endpoint URL. | |
bearer_token | Text | no | Optional bearer token. | |
headers_json | Text | no | {} | Optional extra HTTP headers as a JSON object. |
protocol_version | Text | no | 2025-03-26 | MCP protocol version. |
client_name | Text | no | Agentivity | Client name reported during MCP initialize. |
client_version | Text | no | Client version reported during MCP initialize. | |
timeout_seconds | Number | no | 30 | HTTP timeout in seconds. |
cache_tools | Boolean | no | true | Cache remote tool definitions during the workflow run. |
Plugs into a place that accepts ai.node.tool.*.
AI > Tools > Communication
Section titled “AI > Tools > Communication”AI tools that read and send messages (e-mail).
AI Agent Email Tool
Section titled “AI Agent Email Tool”Resource · in the catalogue: AI > Tools > Communication · type ai.tools.communication.email
Gives an agent a mailbox: search and read e-mails (IMAP), answer or write e-mails (SMTP), and mark or file what it handled. What it may do, and who it may write to, is set on the node.
When to use it. For an agent that works from a mailbox: a support desk answering requests, an assistant sorting what arrived, an agent that sends its result by e-mail. Give only the permissions the job needs.
Effects. reads outside, changes something outside, sends a message.
Fields
| Name | Type | Required | Default | What it is |
|---|---|---|---|---|
inbox | Credentials | no | The mailbox the agent reads (IMAP connection). Needed for reading, answering and filing. Connection of type imap. | |
outbox | Credentials | no | The connection the agent sends with (SMTP). Needed for answering and writing. Connection of type smtp. | |
folder | Text | no | INBOX | The folder the agent works in. |
tool_name_prefix | Text | no | email | Prefix for this instance’s tool names, e.g. “support” → support_search, support_read, support_reply, support_send, support_update. Required when an agent has more than one Email tool connected: tool names must be unique per agent. |
allow_read | Boolean | no | true | Let the agent search and read e-mails. |
allow_send | Boolean | no | true | Let the agent send e-mails (answers and new ones). |
allow_update | Boolean | no | true | Let the agent mark e-mails as read, flag them or move them to a folder. |
allowed_recipients | Text | no | Who the agent may write to: addresses or domains separated by commas (anna@shop.com, @shop.com). Empty = anyone. An answer to an e-mail it read is always allowed. | |
max_body_chars | Number | no | 12000 | The most characters of an e-mail’s text handed to the agent. |
Plugs into a place that accepts ai.node.tool.*.
AI > Tools > Data
Section titled “AI > Tools > Data”AI tools that read and write Data Tables and Vector Stores.
AI Agent Data Table Tool
Section titled “AI Agent Data Table Tool”Resource · in the catalogue: AI > Tools > Data · type ai.tools.data.datatable
Exposes CRUD on a shared Data Table asset as LLM-callable tools.
When to use it. To let an agent search, add, change or delete rows of a Data Table, for example bookings or leads shared by several agents. Turn off what the agent must not do; questions in SQL (totals, averages) stay off unless you turn them on.
Effects. writes in the workspace.
Fields
| Name | Type | Required | Default | What it is |
|---|---|---|---|---|
table_id | AssetRef | yes | The shared Data Table asset this tool reads and writes. | |
tool_name_prefix | Text | no | datatable | Prefix for this instance’s tool names, e.g. “trips” → trips_select, trips_insert, trips_update, trips_delete, trips_find. Required whenever an agent has more than one Data Table tool connected — tool names must be unique per agent. |
max_rows | Number | no | 100 | The most rows one select gives the agent (default 100; 0 for no limit). When a search finds more, the agent is told how many there are and how to get the next ones. |
allow_read | Boolean | no | true | Expose reading rows (select/find) to the agent. |
allow_insert | Boolean | no | true | Expose inserting new rows to the agent. |
allow_update | Boolean | no | true | Expose updating existing rows to the agent. |
allow_delete | Boolean | no | true | Expose deleting rows to the agent. |
allow_query | Boolean | no | false | Expose read-only SQL on this table to the agent (one SELECT at a time: totals, groups, averages…). Off unless turned on. |
Plugs into a place that accepts ai.node.tool.*.
AI Agent Vector Store Tool
Section titled “AI Agent Vector Store Tool”Resource · in the catalogue: AI > Tools > Data · type ai.tools.vectorstore
Lets an AI agent search the connected Vector Store by meaning, and keep texts there when allowed.
When to use it. To give an agent a memory it searches by meaning, and fills itself when storing is allowed: notes on customers, what it learned, answers it found. Connect a Vector Store to it (In-Memory lasts one run; Qdrant, Pinecone or pgvector keep the texts). For what an agent should know from documents, pages or feeds, use a Knowledge Base instead.
Effects. writes in the workspace.
Fields
| Name | Type | Required | Default | What it is |
|---|---|---|---|---|
name | Text | yes | documents | Names the agent’s tools: “Customer notes” gives search_customer_notes (and store_customer_notes). |
description | Text | no | What the texts are, said to the agent so it knows when to search them. | |
limit | Number | no | 4 | The most texts one search gives. |
allow_storing | Boolean | no | false | Also let the agent keep texts in the Vector Store. |
Places for resources
| Place | Required | Accepts | How many |
|---|---|---|---|
| Vector Store | yes | ai.vectorstore.* | 1 |
Plugs into a place that accepts ai.node.tool.*.
AI > Tools > Documents
Section titled “AI > Tools > Documents”AI tools that read documents (PDF, Word, text).
AI Agent Read Document Tool
Section titled “AI Agent Read Document Tool”Resource · in the catalogue: AI > Tools > Documents · type ai.tools.documents.read
Lets an AI agent read a document the user uploaded in the chat (PDF, Word, plain text) as Markdown.
When to use it. To let an agent read a file the user drops in the chat (PDF, Word or plain text) and work from its content: a contract to summarize, a CV to screen, an invoice to check. A scanned page with no text in it cannot be read.
Plugs into a place that accepts ai.node.tool.*.
AI > Tools > Scrappers
Section titled “AI > Tools > Scrappers”AI tools for provider-based web scraping.
AI Agent Web Scraper Tool (Direct)
Section titled “AI Agent Web Scraper Tool (Direct)”Resource · in the catalogue: AI > Tools > Scrappers · type ai.tools.web.scraper.direct
Exposes the direct web scraper to AI agents as a callable tool.
When to use it. To let an agent read a web page from its address and get its main content as clean text, its title and description, and the links to other pages of the same site. Free and needs no account; a page that only shows its content once scripts have run may come back nearly empty.
Effects. reads outside.
Fields
| Name | Type | Required | Default | What it is |
|---|---|---|---|---|
default_render_javascript | Boolean | no | false | Default JavaScript rendering behavior. |
default_extract_markdown | Boolean | no | true | Default markdown extraction behavior. |
default_extract_metadata | Boolean | no | true | Default metadata extraction behavior. |
default_include_html | Boolean | no | false | Return the page’s raw HTML to the agent as well. Off by default: the cleaned markdown and structured data hold what the agent needs, and raw HTML can fill its context with up to 50 KB of markup per call. |
default_timeout_seconds | Number | no | 30 | Default timeout in seconds. |
default_max_content_length | Number | no | 50000 | Default maximum content length. |
rate_limit_profile | Choice | no | medium | Rate limiting and retry profile. Choices: none, fast, medium, slow. |
Plugs into a place that accepts ai.node.tool.*.
AI Agent Web Scraper Tool (Firecrawl)
Section titled “AI Agent Web Scraper Tool (Firecrawl)”Resource · in the catalogue: AI > Tools > Scrappers · type ai.tools.web.scraper.firecrawl
Exposes the Firecrawl web scraper to AI agents as a callable tool.
When to use it. To let an agent read a web page from its address through Firecrawl, a paid service that returns the page’s main content cleaned up. Needs a Firecrawl API key: try the free Direct scraper first.
Effects. reads outside.
Fields
| Name | Type | Required | Default | What it is |
|---|---|---|---|---|
credentials | Credentials | yes | Firecrawl API credentials. Connection of type firecrawl. | |
default_render_javascript | Boolean | no | false | Default JavaScript rendering behavior. |
default_extract_markdown | Boolean | no | true | Default markdown extraction behavior. |
default_extract_metadata | Boolean | no | true | Default metadata extraction behavior. |
default_include_html | Boolean | no | false | Return the page’s raw HTML to the agent as well. Off by default: the cleaned markdown and structured data hold what the agent needs, and raw HTML can fill its context with up to 50 KB of markup per call. |
default_timeout_seconds | Number | no | 30 | Default timeout in seconds. |
default_max_content_length | Number | no | 50000 | Default maximum content length. |
rate_limit_profile | Choice | no | medium | Rate limiting and retry profile. Choices: none, fast, medium, slow. |
Plugs into a place that accepts ai.node.tool.*.
AI Agent Web Scraper Tool (Jina)
Section titled “AI Agent Web Scraper Tool (Jina)”Resource · in the catalogue: AI > Tools > Scrappers · type ai.tools.web.scraper.jina
Exposes the Jina Reader web scraper to AI agents as a callable tool.
When to use it. To let an agent read a web page from its address through the Jina Reader service, which returns the page as clean text. Needs no account, but the page address is sent to Jina: for a page the Direct scraper reads well, prefer Direct.
Effects. reads outside.
Fields
| Name | Type | Required | Default | What it is |
|---|---|---|---|---|
default_render_javascript | Boolean | no | false | Default JavaScript rendering behavior. |
default_extract_markdown | Boolean | no | true | Default markdown extraction behavior. |
default_extract_metadata | Boolean | no | true | Default metadata extraction behavior. |
default_include_html | Boolean | no | false | Return the page’s raw HTML to the agent as well. Off by default: the cleaned markdown and structured data hold what the agent needs, and raw HTML can fill its context with up to 50 KB of markup per call. |
default_timeout_seconds | Number | no | 30 | Default timeout in seconds. |
default_max_content_length | Number | no | 50000 | Default maximum content length. |
rate_limit_profile | Choice | no | medium | Rate limiting and retry profile. Choices: none, fast, medium, slow. |
Plugs into a place that accepts ai.node.tool.*.
AI > Tools > Search
Section titled “AI > Tools > Search”AI tools for provider-based web search.
AI Agent Web Search Tool (SearXNG)
Section titled “AI Agent Web Search Tool (SearXNG)”Resource · in the catalogue: AI > Tools > Search · type ai.tools.web.search.searxng
Exposes SearXNG web search to AI agents as a callable tool.
When to use it. To let an agent search the web and get a list of results (title, address, short extract) from a SearXNG server you run yourself; its address goes in the credentials. To read the pages found, also give the agent an AI Agent Web Scraper Tool.
Effects. reads outside, costs.
Fields
| Name | Type | Required | Default | What it is |
|---|---|---|---|---|
credentials | Credentials | yes | SearXNG credentials. Connection of type searxng. | |
default_top_k | Number | no | 5 | Default maximum number of results. |
default_language | Text | no | Default language hint. | |
default_site | Text | no | Default site/domain filter. | |
rate_limit_profile | Choice | no | medium | Rate limiting and retry profile. Choices: none, fast, medium, slow. |
Plugs into a place that accepts ai.node.tool.*.
AI Agent Web Search Tool (SerpApi)
Section titled “AI Agent Web Search Tool (SerpApi)”Resource · in the catalogue: AI > Tools > Search · type ai.tools.web.search.serpapi
Exposes SerpApi web search to AI agents as a callable tool.
When to use it. To let an agent search the web and get a list of results (title, address, short extract) through SerpApi, a paid service that needs a SerpApi API key. To read the pages found, also give the agent an AI Agent Web Scraper Tool.
Effects. reads outside, costs.
Fields
| Name | Type | Required | Default | What it is |
|---|---|---|---|---|
credentials | Credentials | yes | SerpApi credentials. Connection of type serpapi. | |
default_top_k | Number | no | 5 | Default maximum number of results. |
default_language | Text | no | Default language hint. | |
default_site | Text | no | Default site/domain filter. | |
rate_limit_profile | Choice | no | medium | Rate limiting and retry profile. Choices: none, fast, medium, slow. |
Plugs into a place that accepts ai.node.tool.*.
AI Agent Web Search Tool (Tavily)
Section titled “AI Agent Web Search Tool (Tavily)”Resource · in the catalogue: AI > Tools > Search · type ai.tools.web.search.tavily
Exposes Tavily web search to AI agents as a callable tool.
When to use it. To let an agent search the web and get a list of results (title, address, short extract) through Tavily, a paid service that needs a Tavily API key. To read the pages found, also give the agent an AI Agent Web Scraper Tool.
Effects. reads outside, costs.
Fields
| Name | Type | Required | Default | What it is |
|---|---|---|---|---|
credentials | Credentials | yes | Tavily API credentials. Connection of type tavily. | |
default_top_k | Number | no | 5 | Default maximum number of results. |
default_language | Text | no | Default language hint. | |
default_site | Text | no | Default site/domain filter. | |
rate_limit_profile | Choice | no | medium | Rate limiting and retry profile. Choices: none, fast, medium, slow. |
Plugs into a place that accepts ai.node.tool.*.
AI > Tools > Utils
Section titled “AI > Tools > Utils”Small AI utility tools.
AI Agent Calculator Tool
Section titled “AI Agent Calculator Tool”Resource · in the catalogue: AI > Tools > Utils · type ai.tools.utils.calculator
Evaluates arithmetic expressions for AI agent tool calling.
When to use it. To let an agent get exact results for plain arithmetic (totals, percentages, prices with tax) instead of working them out itself. Not for longer logic or reshaping data: use the JavaScript Tool instead.
Plugs into a place that accepts ai.node.tool.*.
AI Agent Date & Time Tool
Section titled “AI Agent Date & Time Tool”Resource · in the catalogue: AI > Tools > Utils · type ai.tools.utils.datetime
Returns the current date/time in UTC or local time for AI agent tool calling.
When to use it. To let an agent know today’s date and the current time, for example to work out a deadline, a delay or the age of a request. Give it to any agent that has to reason about dates.
Plugs into a place that accepts ai.node.tool.*.
AI Agent Weather Tool
Section titled “AI Agent Weather Tool”Resource · in the catalogue: AI > Tools > Utils · type ai.tools.utils.weather
Retrieves the current weather for a location using the free Open-Meteo API.
When to use it. To let an agent check the weather right now in a city, for example to suggest activities on a trip. It gives the current weather only, not a forecast, and needs no account.
Effects. reads outside.
Plugs into a place that accepts ai.node.tool.*.