The foundation behind Barn Owl. Available to your agents, too.

Hermetiq MCP exposes build and infrastructure tools and data. Barn Owl builds on that foundation with engineered prompts, model routing, and context tailored to your task. Connect your own MCP-compatible agent when you want Hermetiq in your existing workflow.

From a failed build to an evidence-backed change.

  • Query builds, actions and failures
  • Connect cache and infrastructure evidence
  • Apply fixes through your coding workflow
  • Rerun affected work and inspect the result

Let the agent retrieve the same build record an engineer would inspect.

MCP provides a structured interface to Hermetiq data. The quality of an answer still depends on project access, connected telemetry, and engineer validation.

Latest failure

What failed in the latest recorded build, target, or test, and which evidence supports the answer?

Invocation change

Which recorded inputs, outcomes, timings, or execution conditions differ between two builds?

Cache and RBE context

What cache or remote-execution trend is visible in the connected telemetry?

Data gaps

Which source is missing before the agent can answer the question defensibly?

Connect a project, investigate the build, then validate the fix.

Available tools and results depend on the Hermetiq deployment, project permissions, and telemetry connected for that project.

01

Create a project and connect BEP

Establish a real Bazel build record in Hermetiq before asking an agent to investigate it.

02

Configure an MCP client

Connect an authenticated Hermetiq MCP endpoint to a compatible client such as Codex or Claude.

03

Apply and validate

Review the evidence and proposed change. In a coding client with repository and terminal access, let the agent apply the approved fix, rerun affected work, and inspect the result.

Move from an agent question to the underlying build evidence.

A remote action fails because a BUILD rule omits an input. Hermetiq supplies the action evidence, Codex applies a one-line patch, and Bazel verifies the fix. Follow the recorded local sample with its source, logs, MCP responses and reruns.

Read the real build-fix walkthrough →

Invocation comparison product view

An example of the dashboard’s invocation comparison. The walkthrough above contains the actual logs, MCP responses and patch from the recorded local build.

Keep the model-facing interface bounded by real project data.

Structured interaction

MCP standardizes discovery and invocation of the tools and resources exposed by a configured server.

Project-scoped results

Hermetiq responses are bounded by authenticated project access and the telemetry retained for that project.

Source-backed answers

Build explanations should point back to the invocation, target, test, profile, action, or infrastructure evidence used.

Client-controlled changes

Code edits and validation commands use the coding client’s repository access, command execution, permissions and approval workflow.

Verify the underlying telemetry and performance model.

Connect a real project, then give your agent build evidence.

Start with one Bazel invocation so every MCP-assisted investigation has an evidence boundary.

Technical review: September 8, 2026 · Verify configuration details against the Bazel and Buildbarn versions deployed in your environment.