Skip to content

AI, assistant and MCP

Toolora exposes authorized business context to its assistant and public MCP clients. Current tool contracts describe arguments, results and supported proposals. Context selection and permissions are enforced before content is used. MCP authorization and explicit human confirmation constrain the operations available to a client.

The public MCP contract and generated catalogue describe available tools. The connected server’s tools/list is authoritative for that connection. Structured results include source references, freshness, missing information and truncation. Payment conclusions require the relevant reconciliation coverage. The contractor-matching gateway is a separate interface with a different purpose.

MCP tools and responses describes discovery, evidence envelopes, continuations, errors and effect-bearing tool families.

Context builders select information according to caller permissions and the intended audience. Source text remains data rather than instructions. Evidence references identify the records or originals used in a result. Document interpretation preserves source identity so a reviewer can distinguish an extracted proposal from a confirmed business fact.

Context and source evidence explains selection before model use, response budgets and source-review limits.

An agent draft is a saved proposal with its own revisions. Human confirmation applies the reviewed version through a defined operation. A confirmed quote proposal creates an editable quote; it does not send the quote or record customer acceptance. Quote acceptance has a separate actor, document version and legal meaning.

Proposals and human confirmation compares quote, invitation, document and import-supplement confirmation boundaries.

AI usage and tool lifecycle events have distinct logging responsibilities. The MCP contract includes bounded results and caller-scoped access. Unsupported capability requests enter the product-feedback workflow rather than granting new authority. Logging rules must protect business text and credentials while retaining enough operation identity for diagnosis.

Usage, audit and feedback describes usage records, invocation metadata, separate budgets and unsupported requests.