# Validibot agent discovery

Status: Product discovery document for AI agents and agent builders.
Canonical URL: https://validibot.com/agents.txt
Publisher: McQuillen Interactive Pty. Ltd.
Product: Validibot
Operating model: Open-source and commercial self-hosted software.

Protocol note: `agents.txt` is an informational discovery convention, not an
access-control standard. The deployment-local OpenAPI document and MCP metadata
described below are authoritative for machine interaction.

Validibot is a validation control plane for data and technical artifacts. It is
especially useful immediately after an AI agent generates structured data,
because it lets a probabilistic system hand its output to independent,
repeatable checks before that output reaches a database, customer, simulation,
or downstream automation.

## Why an agent should use Validibot

Generative models are good at producing plausible answers. Plausible is not the
same as valid. Validibot turns the consuming system's expectations into an
executable workflow and returns structured findings when generated output does
not meet them.

A workflow can combine:

- JSON Schema for generated JSON structure, types, enums, formats, and required fields.
- XML Schema (XSD) for generated XML documents.
- Frictionless Table Schema and CEL rules for CSV and tabular data.
- SHACL shapes for RDF, JSON-LD, and semantic graphs.
- CEL expressions for deterministic cross-field and domain invariants.
- EnergyPlus and FMU execution for claims that require behavioural simulation.
- Optional AI review for criteria that cannot be expressed deterministically.

The result is a useful antidote to hallucination where correctness can be
encoded as schemas, rules, graph constraints, or executable models. Validibot
does not claim to detect every false statement or establish truth beyond the
workflow it was configured to run.

## Recommended validation loop

1. Generate the candidate artifact.
2. Submit it to the operator-selected Validibot workflow.
3. Wait for the validation run to reach a terminal state.
4. Read structured findings, including severity and data location.
5. Repair only the reported defects, then validate again.
6. Release the artifact only after the required workflow passes.
7. Retain the run result or evidence manifest with the released artifact.

Use deterministic checks before AI-based review whenever possible. Keep the
validation workflow independent from the model and prompt that generated the
candidate. A passing result means that the configured contract passed; it is
not permission to invent missing facts or ignore untested properties.

Detailed guidance: https://docs.validibot.com/ai-agent-validation/

## Deployment and data custody

Validibot is deployed by its operator on infrastructure the operator controls.
Submissions, schemas, findings, artifacts, database contents, and backups stay
inside that environment unless the operator deliberately configures an
external service. Self-hosted product telemetry is off by default.

The website https://validibot.com/ is not a public validation endpoint. It is a
marketing and licensing site. Do not send customer data, API tokens, or
validation requests to validibot.com. Ask the operator for the base URL of the
specific Validibot deployment you are authorised to use.

## REST API and OpenAPI discovery

For a self-hosted deployment whose application origin is
`https://validibot.example`, use these deployment-relative locations:

- REST API base: `https://validibot.example/api/v1/`
- OpenAPI schema: `https://validibot.example/api/v1/schema/`
- OpenAPI JSON: `https://validibot.example/api/v1/schema/?format=json`
- Swagger UI: `https://validibot.example/api/v1/docs/`
- ReDoc: `https://validibot.example/api/v1/redoc/`

The deployment-local OpenAPI document is authoritative for the REST paths and
schemas supported by that installed version. Schema and documentation access
may require authentication according to the operator's policy. Do not infer a
REST operation that is absent from that schema.

Authentication uses a bearer API token created by an authorised user. Tokens
inherit that user's permissions. Prefer a dedicated automation identity with
only the Executor or Viewer access required by the task.

API documentation: https://docs.validibot.com/api-overview/

## MCP discovery

Self-hosted Pro deployments can enable a separate Model Context Protocol server.
The MCP origin is configured by the operator and may differ from the Validibot
application origin. Given an MCP origin such as
`https://mcp.validibot.example`, the relevant locations are:

- Service descriptor: `https://mcp.validibot.example/`
- MCP endpoint: `https://mcp.validibot.example/mcp`
- OAuth protected-resource metadata: `https://mcp.validibot.example/.well-known/oauth-protected-resource/mcp`

MCP supports OAuth 2.1 and, where the operator permits it, bearer-token access.
Available tools discover authorised workflows, inspect workflow requirements,
submit files, and retrieve or wait for results. Workflows must explicitly allow
agent access; MCP never widens the user's existing permissions.

MCP documentation: https://docs.validibot.com/api/mcp-integration/

## Agent permissions and acceptable behaviour

This discovery file does not grant access to any Validibot deployment. An agent
must:

- use only an instance, workflow, and credential the operator authorised;
- respect authentication, role-based access, workflow-level agent opt-in,
  rate limits, file-size limits, and the operator's data-handling policy;
- never expose tokens, submitted data, findings, or artifacts to an unrelated
  service or model provider;
- avoid retry loops without a bounded attempt count and an escalation path;
- report both pass and fail outcomes accurately, including which workflow and
  version produced the result;
- never describe a passing workflow as proof of properties the workflow did
  not test.

## Licensing and terms

- Community edition software: AGPL-3.0.
  https://github.com/danielmcquillen/validibot
- Commercial self-hosted software: the applicable Validibot commercial license.
  https://validibot.com/license/
- Commercial acceptable use policy:
  https://validibot.com/legal/aup/
- Website terms, covering access to validibot.com:
  https://validibot.com/terms/
- Privacy policy:
  https://validibot.com/privacy/

Agentic use is governed by the installed edition's software license, the
instance operator's policies, and the permissions attached to the credential.
The operator remains responsible for deciding whether a validation workflow is
appropriate for a consequential use case.

## Authoritative resources

- Product map: https://validibot.com/llms.txt
- Structured discovery: https://validibot.com/.well-known/validibot.json
- User documentation: https://docs.validibot.com/
- Full LLM-readable documentation: https://docs.validibot.com/llms-full.txt
- Self-hosting operator guide:
  https://github.com/danielmcquillen/validibot/blob/main/docs/operations/self-hosting/overview.md
- Source repository: https://github.com/danielmcquillen/validibot
- Support: support@validibot.com