AI agents & MCP

Agents build beside your engineers.

Assign tasks to AI agents as you would to a colleague. They create data, comment, move workflows forward and hand work to other agents, within the rights you give them. Nothing is merged without an engineer's approval.

−50%
time spent on repetitive engineering tasks
−30%
time to market
AI mechanical engineerService account · arm program

Task

The bracket moves to supplier B: +180 g. Update the mass budget, flag what is at risk and draft the change request.

Plan

  1. Read the BOM and the bracket's links
  2. Run the mass roll-up in Python: arm 4.82 → 5.00 kg
  3. Flag REQ-112 (mass budget) and 3 tests to rerun
  4. Open branch auto/bracket-b and draft CR-118

Output

CR-118Bracket supplier change
  • Arm assembly mass4.82 → 5.00 kg
  • REQ-112 mass budgetmargin −2 %
  • Tests to rerun3
  • Evidenceattached
Waiting for your approval
Every action on the item's activitySimulated example

01

Skills, automations and AI engineers.

Three building blocks turn the routine checks of your team into work that agents do for you. Engineers spend half as much time on repetitive tasks and keep their week for design decisions.

Skills

Skills describe your standards and team conventions in plain language: naming rules, review criteria, clauses of a standard. Agents apply them to every item they create or check.

Automations

A pull request, a CAD revision or a new supplier PDF starts the agent workflow you assigned to it. The checks your process requires run on every change, without anyone launching them.

AI engineers

Agents named after an engineering job take the tasks your team assigns them: tracing dependencies, checking test coverage, updating roll-ups and drafting change requests.

02

A team of AI engineers, one per engineering job.

You assign tasks to each agent as to a team member. It applies your skills, can hand part of the work to another agent, and submits the result to an engineer for approval.

AI engineering system assistant

Shows the parts, requirements and tests a change affects before you approve it.

AI mechanical engineer

Keeps the mass and cost roll-ups of the BOM up to date after every change.

AI requirements engineer

Links each requirement to the design element and the test that prove it, and flags the gaps.

AI verification engineer

Lists the tests a change invalidates and reopens them for a rerun.

AI release engineer

Prepares the baseline and the evidence for each design review gate.

Your own agent

Runs your own instructions with your skills and the rights of its service account.

03

Ask what if. Get the full analysis.

  1. 1

    Ask in plain words

    What if the bracket moves to supplier B?

  2. 2

    The agent works on a branch

    It follows the links from the bracket across the system and computes the answer on a branch, without touching the baseline.

  3. 3

    A decision report comes back

    The report lists the affected items, the changes they need and whether the requirements still hold.

  4. 4

    Before and after, side by side

    You compare both versions, mass and cost included, and decide.

What if the bracket moves to supplier B?
Decision reportTodayIf changed
  • Arm assembly mass4.82 kg5.00 kg
  • Mass budget margin+1.6 %−2.0 %
  • Requirements affected01
  • Tests to rerun03
  • Variants touched02
Feasible with a lighter fastener setSimulated example

04

Every change is checked by an agent.

When a change comes in, the agent finds the parts, requirements and tests it affects, proposes the updates on a branch and notifies the reviewers who must approve them.

01

A change comes in

GitHub PR #42 · firmware
Supplier PDF · bracket rev B
CAD revision · arm v3

02

The agent finds what it affects

REQ-112 mass budget at risk
TEST fatigue to rerun
Variant XL touched

03

Fixes proposed on branches

auto/arm-rev-c
auto/mass-budget-v4
CR-118 drafted

04

Engineers notified, the decision stays human

Reviewers notified
Lead mechanical · approved
Quality · reviewing
Nothing merges into the baseline before sign-off Simulated example

05

Agents take the tasks you assign. Engineers approve their work.

  • You assign the task

    You assign it like any task, or a trigger does it for you: a pull request, a supplier PDF or an updated standard.

  • The agent works on a branch

    It creates the missing items, comments on what it found, moves the workflow forward and hands sub-tasks to other agents, without touching the baseline.

  • An engineer reviews and merges

    The branch is merged into the baseline only after a reviewer approves it, like a pull request.

Koddex · Braking system · live
You AI systems engineer
AI systems engineerYou
Live activity
    Test coverage0 %
    Waiting for your sign-off
    Simulated run

    06

    Write your engineering standards once.

    • Plain instructions, no code

      You write naming rules, quality criteria and the clauses of your standards as plain text.

    • Reused by every agent

      Every agent you run applies the same rules, the same way, on every item.

    • Combined in workflows

      A compliance check can apply your verification rules and your naming rules in a single run.

    AI verification engineer

    Agent settings · Koddex

    Configuring its skills
    • Structure a customer specification
    • Allocate requirements to the system
    • Write test skeletons from requirements
    • Run impact analysis on every change
    • Draft change requests for review
    Simulated run

    07

    Bring your own agent.

    Connect Claude, ChatGPT or your own agent to Koddex through MCP. It reads and writes your engineering data with the rights of the account that connects it, and every action is logged.

    Claude · connected to Koddex through MCP

    08

    Also built into Koddex agents.

    Sandboxed Python

    Agents run mass roll-ups, tolerance stack-ups and unit conversions in Python, in an isolated sandbox. The results feed the agent's reasoning and appear in its transcript.

    Branches and merge

    Each agent fix lives on its own branch, which you compare with the baseline and merge only after approval.

    Values computed by Koddex

    Mass, cost and coverage roll-ups are computed by Koddex. The agent reads these values and never makes them up.

    Service accounts with access tags

    Each agent connects with a service account whose access tags decide what it can see and edit, down to a single item.

    Every action on the activity

    Each action is recorded on the item's activity: who, when and with which data. You review it like any other change.

    The model you choose

    Connect Claude, ChatGPT or your own agent to Koddex through MCP.

    A 20-min demo. Your first use case live in 3 days.

    Bring one question your team asks every week. In the demo, we ask a Koddex agent the same kind of question on a product like yours.

    We use cookies to keep the site running and to understand how it's used so we can improve it, like most sites do. Read the cookie policy