How to Connect the Ethira MCP to Claude and Manage Your Inventory

TL;DR

• Connect Claude.ai to your Ethira workspace as a custom connector in under five minutes.
• Use plain English in Claude to list, create, and update systems, products, datasets, microservices, devices, physical locations, processes, and teams.
• Claude inherits your workspace permissions — nothing it does in Ethira bypasses RBAC, and every action is recorded in the workspace activity log.
• Keeping inventory current is a prerequisite for DORA, the EU AI Act, and ISO 42001. Letting an agent do the data entry is how you actually keep it current.

What the Ethira MCP server gives you

The Ethira MCP server is the same workspace you use in the browser, exposed to any AI client that speaks the Model Context Protocol. For asset inventory specifically, the server exposes tools and resources for every inventory entity in Ethira:

• Systems — applications, platforms, internal tools
• Products — the things your company sells or operates
• Microservices — individual services and their owners
• Datasets and datapoints — the data your systems hold, with personal-data classifications
• Devices — laptops, servers, IoT
• Physical locations — offices, data centres
• Processes — business processes that connect systems
• Teams — who owns what
• Source code repositories — linked to systems and microservices

Each entity has the standard list_, get_, create_, update_, and delete_ tools, plus ambient ethira://inventory/ resources Claude can read without a tool call. The server is scoped to the workspace tied to your account — you cannot accidentally write to another tenant.

Step 1 — Open the MCP page in Ethira

In the Ethira app, go to Settings → MCP Server.

That page does two things:

1. Shows you the connector URL to paste into Claude — for production workspaces, this is https://api.ethira.dev/v1/mcp.
2. Has an Open Claude connectors button that takes you straight to the right page in Claude.ai so you don't have to hunt for it.

You don't need to generate an API key for the Claude.ai web client. Claude uses OAuth — it will bounce you back to Ethira to sign in and pick a workspace, and the connector inherits your own permissions from there.

Step 2 — Add Ethira as a custom connector in Claude.ai

In Claude.ai:

1. Open Settings → Connectors.
2. Click Add custom connector.
3. Paste the values from the Ethira MCP page:
   • Name: Ethira
   • URL: https://api.ethira.dev/v1/mcp
4. Click Add.

Claude will redirect you to app.ethira.dev to sign in. Use the same credentials (or SSO) you use for the Ethira app itself. Pick the workspace you want to connect, approve the consent screen, and Claude will send you back to the connector list with Ethira marked as connected.

On Claude Team and Enterprise, only workspace Owners can add the connector. Owners do it from Admin settings → Connectors → Add custom connector, paste the Name and URL above, and click Add. Individual members then go to their own Settings → Connectors and click Connect next to Ethira to sign in. On Free, Pro, and Max, any user can add the connector directly from Settings → Connectors — no admin step required.

If anything fails on this step — usually it's a wrong workspace pick or a cancelled consent screen — start the flow over from Settings → Connectors → Add custom connector. Workspace admins can also check Settings → Activity Log for oauth.client_registered and oauth.access_granted entries to confirm Claude reached the workspace.

Step 3 — Your first inventory queries

Open a new conversation in Claude.ai. Click the + button in the lower-left of the composer (or type /), hover over Connectors, and make sure Ethira is toggled on. From the same menu you can also pick Tool access → Always available if you want Claude to load Ethira at the start of every turn instead of auto-deciding. Then try the prompts below — each one is copy-pasteable.

Read what's already there

"List all systems in my Ethira workspace, sorted by criticality. Group them by the team that owns them."

"Show me every dataset tagged as containing personal data, and tell me which system holds each one."

"How many microservices do we have without a named owner?"

Claude will call list_system, list_dataset, list_microservice, follow the relations into teams and personal-data categories, and answer in a single turn. For broad questions, Claude often reads the ethira://inventory/summary resource first instead of calling a tool — it's faster.

Add new inventory

"Add a new microservice called billing-webhooks. It's owned by the Payments team, runs on AWS, and processes Stripe events. Link it to the Stripe vendor and the Billing system."

"We just opened a new office in Lisbon. Add it as a physical location, and link the Marketing team to it."

"Register a new dataset called customer_support_transcripts. It contains personal data (names, email addresses, and free-text customer messages). It's held by the Zendesk system."

Claude will call the right create_ tools, fill in the relations, and confirm what it created with the IDs so you can audit the change.

Keep things current

"Find every microservice owned by the 'Platform' team and update its environment to 'production' if it has a -prod suffix in its name."

"List all devices last reviewed more than six months ago. Mark them for a refresh review."

"Remove the link between the reporting system and the legacy-warehouse dataset — we decommissioned it last month."

Every create_, update_, and delete_ action shows up in Settings → Activity Log with the connector that triggered it. You always have a paper trail.

Why this matters

The reason most inventories rot is not that nobody cares — it's that the people closest to the truth (the engineers who shipped a new service, the ops manager who opened a new office, the data lead who added a new dataset) do not have time to log into a GRC tool and fill in a form.

Letting them update inventory by talking to Claude in the tool they already use removes that friction. The agent enforces the schema, picks up the relations, and writes back to Ethira under their own identity, so nothing bypasses RBAC and every action is auditable.

That matters because every meaningful AI and operational-resilience regulation now expects a current inventory:

• DORA requires a live Register of Information for every critical ICT third party — which depends on knowing which of your systems and microservices use them.
• The EU AI Act requires documented oversight of AI systems used in high-risk contexts, which starts with knowing those systems exist.
• ISO 42001 explicitly requires a documented inventory of AI systems in scope of the management system.
• GDPR Article 30 RoPA records reference the systems, datasets, and personal-data categories you maintain in inventory.

A stale inventory is a compliance gap. An agent that keeps it fresh in the background is a control.

What to try next

Once your inventory flow is comfortable, the same connector unlocks the rest of the Ethira surface:

• Vendors and contractual relationships — list_vendor, create_vendor, update_contractual_relationship for TPRM in Claude
• Record of Processing Activities — list_/get_/create_/update_processing_activity to keep GDPR Article 30 records current
• Risks and findings — create_risk, list_finding to capture risk observations as they come up in conversation
• Documents — semantic search across your uploaded policies and DPAs

If you want to lock down what Claude can do (for example, give it read-only access while you trial it), reissue the MCP token with a narrower scope from Settings → API Keys and add it to the connector. Permissions are the same RBAC system as the app — Claude cannot do anything you couldn't do yourself.

Frequently asked questions

Do I need a separate API key for Claude.ai? No. The Claude.ai web client uses OAuth — it sends you to Ethira to sign in and inherits your permissions from that session. You only need a long-lived MCP token for non-OAuth clients (Cursor, VS Code, custom scripts).

Can Claude see data from other workspaces? No. The OAuth flow forces you to pick one workspace at connect time. The token Claude receives is bound to that workspace — every tool call is automatically scoped and there is no parameter to escape it.

What happens when I leave the company? When your Ethira account is disabled, the OAuth session backing the Claude connector is revoked immediately. Every tool call after that fails with an authentication error.

Is every change Claude makes audited? Yes. Every create_, update_, and delete_ call goes through the same activity-logging pipeline as the web app. Workspace admins can see exactly which connector and which user made every change in Settings → Activity Log.

Can I roll back something Claude got wrong? For most entities, yes — the activity log records the before/after of every change, and inventory entities support delete_ and re-create. For destructive operations, ask Claude to summarise the planned change before executing it ("show me what you'd do, but don't run it yet").

Is this an official Anthropic-supported integration? Custom connectors using remote MCP are Anthropic's official integration mechanism for third-party services on Claude.ai, currently in beta across Free, Pro, Max, Team, and Enterprise plans. Ethira's MCP server implements the public Anthropic MCP specification, so it works the same way as any other custom connector — there is no Ethira-specific Claude integration to maintain, and the same connector keeps working as Anthropic graduates the feature out of beta.

Sources

1. Model Context Protocol specification — modelcontextprotocol.io
2. Get started with custom connectors using remote MCP — Anthropic Help Center
3. EU Digital Operational Resilience Act (DORA), Regulation (EU) 2022/2554 — EUR-Lex
4. EU Artificial Intelligence Act, Regulation (EU) 2024/1689 — EUR-Lex
5. ISO/IEC 42001:2023 — Information technology — Artificial intelligence — Management system — ISO