Search the web

Let the LLM search the web, through a Web search connection, for pages no app table can answer about. Each result is a title, a URL and a snippet; pair it with Fetch web page so the model can read a result it picks.

Ingredient Family: llmQueryToolIngredients

Outline Display: Allow LLM to search the web ({{toolName}}) through {{Web search}}: {{toolDescription}}
Results per call: {{maxResults}}{{domains:
Domains: {{domains}}
Listed domains are blocked: {{domainsBlocked}}}}
Requires approval: {{requiresApproval}}{{advisory:
Not enforced by the server: {{advisory}}}}

Details

The search runs through the Web search api named in the settings – a Brave Search api holding a subscription key – so it works with every LLM provider, including ones whose own built-in search cannot be combined with tools. The model sends a query and, optionally, how many results it wants; the tool answers with up to maxResults of them, each a title, url and snippet. Nothing is fetched: a result’s page is read with a Fetch web page tool, if the module offers one. domains is a comma-separated list of domain names, each covering the domain itself and every host under it (example.com covers docs.example.com): the only domains results may come from, or, with domainsBlocked: true, the domains whose results are dropped. A result outside the policy is dropped before the model sees it; when the policy can drop results the engine is asked for its whole page and the first maxResults survivors are answered, and a sole allowed domain is also pushed into the search itself as a site: restriction, so the engine spends its results on pages the model may use. An entry that names no host – a scheme the app may not reach, say – fails every call, naming the entry, until the setting is corrected. Every call is audited under the tool’s wire name with the query as its arguments, whatever its outcome. A key the engine rejects fails the call as unauthorized, a quota run dry or an engine that cannot be reached as a malfunction, and a connection with no key at all as not connected; the model is told which, and the message names the fix.

Settings

Required

  • toolName : constant value (type: text; required, non-nullable)
    The name the LLM calls this tool by: its wire name, which MCP clients, approvals, audit lines and dashboard policy all key on, so renaming it is renaming the tool for every one of them. Required. 1-64 characters of letters, digits, underscore and hyphen, taken exactly as written; a tool whose companions append a suffix to it (a lookup’s _distinct_values, a SQL query’s _schema) must leave room for the suffix under the 64. Two tools in one module cannot share a name.

  • Web search : API connection (API category: Web search)
    The Web search connection the searches go through
    Set this to the name of an api: item whose provider: is in the Web search category, or which reaches Web search on another category’s behalf – a brokered api counts when its own settings point it at that service, and its provider page says which setting decides. Do not set it to a service name; pick the API the service points at.

Optional

  • toolTitle : constant value (type: text; optional)
    The human-readable name an MCP client shows for this tool (MCP’s title), never read by a model. Omitted, it is derived from the wire name: search_open_orders becomes “Search Open Orders”. Two tools in one module cannot share a title, derived or authored.

  • toolDescription : constant value (type: text; required, non-nullable)
    What the LLM should use web search for, to help it decide when to search rather than answer from app data or from memory

  • advisory : constant value (type: text; optional)
    A rule the server does NOT enforce, sent to the model as guidance under an explicit ‘not enforced’ header; use only when no deterministic form (Require, parameter constraints, allowedRecipients…) exists. The checker warns and the dashboard shows it as not guaranteed.

  • requiresApproval : constant value (type: bool; required, non-nullable)
    Require the user to approve each invocation of this tool before it runs. Only honored inside a Chat concept, where the agent pauses mid-turn and shows Approve/Reject buttons; rejecting tells the LLM the call was denied and ends the turn. Setting it true on a tool used by any other agent (e.g. a row-action AskAnLLM triggered by a button) is rejected by the NectryCore typechecker – except under MCP Server, which only warns. When unset, the tool is gated exactly when its steps make it anything but read-only – the same reading its MCP category comes from, so a tool that annotates itself read-only is not gated – and only where a pause is honored: an unattended surface never invents a gate it cannot keep.

  • maxResults : constant value (type: integer; required, non-nullable)
    The most results one call may return, from 1 to 20 (the engine’s own bound; more is read as 20). The model may ask for fewer.

  • domains : constant value (type: text; optional)
    Comma-separated domains, each covering the hosts under it: the only domains results may come from, and one domain alone is also sent to the engine as a site: restriction. An entry may be spelled as a URL, or with a port or path, of which only the host counts. Unset, results may come from any domain.

  • domainsBlocked : constant value (type: bool; required, non-nullable)
    When true, domains lists the domains whose results are dropped instead, and results may come from any other domain.