What is the Koddex MCP server?

An MCP endpoint that exposes Koddex as a set of tools your LLM client can call. The model does not scrape a UI or guess at an API; it invokes typed operations on your engineering graph.

Which LLM clients can connect to it?

Any MCP-capable client. Teams run it from Claude, ChatGPT, Gemini, Mistral AI and Copilot, and the same server serves all of them.

Does the agent see everything in my workspace?

Only what you scope it to. An agent is a principal with read and write rights on specific entities in specific projects, exactly like a human account.

Can an agent change a frozen baseline?

No, unless you grant it that right, and even then the approval workflow applies. The common setup is read-only on released baselines and write on drafts.

How do I know what the agent did?

The audit trail. Every MCP call is recorded with the principal, the timestamp and the permissions it acted under, alongside human edits.

Can it build a whole model from a prompt?

Yes. Describe the assembly and the agent assembles the structure through Koddex tools: sub-components, linked requirements and certifications, created as governed entities rather than text in a chat window.

What are workflows here, exactly?

No-code trigger and action chains that live inside the model. When a requirement is edited, recompute coverage, flag impacted tests and notify the lead, for example.

Does our data leave our perimeter?

The MCP server runs against your workspace under your access rules. What a client can read is what you scoped it to read, and the trail records it.

MCP Koddex

Build your models in minutes, not months.

Describe a working practice to your LLM client and the Koddex MCP server creates it as governed items: entities, attributes, relations and the process around them. Not a transcript you have to retype.

koddex · MCP server

Koddex MCP, live

One MCP server, every client your team already uses

ClaudeAnthropic

ChatGPTOpenAI

GeminiGoogle

Mistral AIMistral

CopilotGitHub

Your LLM client, wired straight into the model.

Connect Claude, ChatGPT, or Gemini to the Koddex MCP server and let them read, build, and edit your engineering data, under the exact same permissions and audit trail as your team.

Koddex MCP server

Build entire models from a prompt.

Ask in plain language and the agent assembles the full structure through Koddex tools: assemblies, sub-components, linked requirements and certifications, created as native governed entities, not text in a chat window.

Works from: Claude, ChatGPT, Gemini, Mistral AI, Copilot

Build your models

claude.ai · koddex-mcp mcp

Create the BOM of a 6-axis robot arm: the full tree with requirements and certifications.

koddex.create_item("6-axis robot arm")

koddex.add_children · Base + Axis 1…6

koddex.link · ISO 10218 · ISO/TS 15066

Bill of materials · tree312 components

6-axis robot arm ISO 10218-1
└Base assembly
└Axis 1 · Shoulder ISO/TS 15066
└Servomotor 750 W
└Harmonic reducer
└Absolute encoder
└Axis 2 · Lower arm
└Servomotor 400 W
└Axis 3–6 + end effector

Agents

Agents that act inside your model.

Spin up agents that read, write, and propose changes on your engineering graph, scoped to the entities they're allowed to touch, fully audited.

See the integrations

koddex · Agents

Impact AnalystComputes downstream impact of any proposed change

Compliance ScannerMaps requirements to ISO 13485, MDR, AS9100 clauses

Test DrafterGenerates test cases from requirement specifications

Workflows

Engineering workflows, no-code.

Trigger anything from anything. "When requirement changes → notify reviewers, recompute coverage, flag impacted tests." All inside the model.

koddex · Workflows

Requirement edited

Recompute coverageFlag impacted testsNotify lead

Baseline frozen

Lock all entitiesHash + archiveExport DHF section

Change request opened

Run impact analysisRoute to reviewersBlock conflicting edits

Build your models in days minutes.

The Koddex MCP server doesn't just read your data, it exposes modeling. From Claude, Gemini or ChatGPT, ask for the BOM of a 6-axis robot arm and the agent builds the cascading assembly tree, hundreds of components and sub-components, with linked requirements and certifications, exactly like on the Custom templates page.

claude.ai · workspace · koddex-mcp mcp connected

In Koddex, create the full BOM of a 6-axis robot arm: generate the cascading assembly tree with every component and sub-component (motors, reducers, encoders, brakes, bearings…), and link the requirements and certifications at each level.

tool ·koddex.create_item("6-axis robot arm")

tool ·koddex.add_children · Base + Axis 1…6 + end effector

tool ·koddex.add_children · 312 components across 6 levels

tool ·koddex.add_attribute · mass, torque, part number…

tool ·koddex.link · 48 requirements → components

tool ·koddex.link · ISO 10218-1 · ISO/TS 15066 · ISO 13849

Done. The assembly tree is live: 312 components across 6 levels, 48 requirements and 12 certifications linked. A change on any component will roll its impact up the whole tree.

Bill of materials · assembly tree312 components

6-axis robot armRBT-ARM-6 ISO 10218-1 CE Machinery
└Base assemblyASM-BASE
└Slew bearingBRG-140
└Cable managementCBL-SYS
└Axis 1 · ShoulderAX-1REQ-014 payload ISO/TS 15066
└Servomotor 750 WMOT-750
└Harmonic reducerHDR-50REQ-031 backlash
└Absolute encoderENC-19b
└Holding brakeBRK-24 ISO 13849 PL d
└Axis 2 · Lower armAX-2REQ-018 reach
└Servomotor 400 WMOT-400
└Harmonic reducerHDR-32
└Axes 3–6 + end effector · 214 more components

Ask Claude to build a bill of materials…

The same custom models and trees you'd build by hand, created by prompt. See Custom templates

Same access model as humans.

An AI agent is just another principal. Give it scoped read/write rights on specific entities, in specific projects, gated by the same approval workflow you use for engineers.

koddex · Permissions

Agent scope

Read-only on Released baselines

Write on draft Requirements (subsystem: Powertrain)

Cannot modify Risk items (ISO 14971)

Must request approval for entities under Baseline v1.2

AI and engineers on the same canvas.

Agents and humans work the live model side by side. Every edit is attributed, permission-checked, and traceable. The AI is just another principal on your engineering graph.

Insulin Pump v3.2 · live

SR-128 · Glucose accuracy

PriorityHigh

VerificationTC-501

StatusIn Design

SR-129 · Battery autonomy

PriorityHigh

VerificationTC-502

StatusReleased

BOM · Piezo pump

Part #PMP-43-A

Revv1.0

SupplierTTP Ventus

Test TC-077

ProcedureStandalone

Pass crit.± 5%

Last runPASS

AI Agent

Marie · Quality

What an MCP server changes about putting AI near engineering data.

The blocker was never whether the model could help. It was whether you could let it near the data without losing the audit trail.

Agents are principals, not integrations

An agent gets scoped read and write rights on specific entities in specific projects, gated by the workflow you already use for engineers.

Same

Access model as a human engineer

Every agent action is attributed

Edits made through the MCP carry an author, a timestamp and the permissions they ran under, so the trail survives the audit.

100%

Agent edits in the audit trail

Models built in a prompt, not a project

Describe an assembly in plain language and the agent creates the entities, attributes and links as native governed items.

Minutes

From prompt to governed model

Frequently asked questions

Bring AI to your engineering without losing control.

See how Koddex makes AI agents safe, governed, and useful inside the engineering loop.