Which AI clients can connect to TransformRadar?

Cursor, Claude, ChatGPT, Mistral Vibe and Google Gemini have public client ids in the catalogue. An Org Admin chooses which of them the organisation allows.

Can a connected agent change my data without me?

No. Every write tool returns a preview, and nothing is applied until the person confirms it. If the record changed in the meantime, the commit is refused.

Does using MCP send my data outside the EU?

The connecting agent’s provider handles the conversation, so what the agent asks for in a session is processed by that provider. TransformRadar’s own hosting and AI stay in the EU. An Org Admin acknowledges this before enabling external access, and it is off by default.

Do I need a developer to set it up?

No. An Org Admin enables access and allows the clients; each user then adds the endpoint and client id in their AI tool and signs in through the browser. The in-product guide shows the exact values for your organisation.

What is the difference between Blip and MCP?

Blip is the assistant inside TransformRadar, with no setup and no data leaving the European chain. MCP lets your own AI client connect from outside. Both use the same governed actions and the same preview-then-confirm protocol.

Is external MCP on by default?

No. It is off until an Org Admin acknowledges the data-egress notice and enables it, and write tiers are off until enabled separately.

For AI agents

Does TransformRadar have an MCP server, and how do I connect an AI agent?

TransformRadar is a Model Context Protocol (MCP) server. An Org Admin switches on access and chooses which AI clients are allowed — Claude, ChatGPT, Cursor, Gemini and Mistral Vibe. Each user connects with a standard sign-in, and the agent acts as that user, through the same governed actions as the product, and every write is a preview a person confirms.

What a connected agent can do

Once connected, an agent sees a set of tools generated from the same action layer the product’s own buttons use. Reads come with access. Writes are switched on separately by the Org Admin, in two tiers, and each one is a preview until a person confirms it.

In practice: ask Claude what slipped on a project this week and which risks are open without an owner; have Cursor turn a design note into delivery items under the right epic; let ChatGPT draft a risk from a meeting transcript, then confirm it in one click.

Reads: Projects, the plan and its tasks, the delivery backlog with cycles and workflow states, the governance log, status reports, steering plans, value realisation and Programme Health.
Writes, task tier: Create and update tasks and delivery items, governance log entries, value drivers, measures and readings; create projects and instantiate modules.
Writes, governance-draft tier: Draft and update the weekly status report and the steering plan — the artefacts a steering group signs off.
Never exposed: Membership, invitations and role changes are not tools. Starting, completing and configuring sprints stay human actions in the product.

How to connect, step by step

External MCP is off by default. Connecting takes one decision by an Org Admin and one sign-in by each user. The in-product setup guide under Organisation settings shows the same steps with your organisation’s live endpoint filled in.

1. An Org Admin enables external MCP: Organisation settings → External MCP. Read and acknowledge the data-egress notice, then enable external access.
2. Allow the clients your organisation uses: Turn on the catalogue entries for Cursor, Claude, ChatGPT, Mistral Vibe or Google Gemini. Their client ids are public and the same for every organisation: transformradar-cursor, transformradar-claude, transformradar-chatgpt, transformradar-mistral-vibe and transformradar-gemini.
3. Optionally enable write tiers: Task-tier writes and governance-draft writes are separate toggles, both off by default. Reads need no toggle beyond access itself.
4. Note the MCP endpoint: The settings page shows the endpoint URL for your deployment. It ends in /mcp with nothing after it.
5. Configure the client: In Cursor, add the endpoint and client id to the MCP configuration file. In Claude, ChatGPT, Gemini or Mistral Vibe, register a custom MCP connector pointing at the endpoint with the matching client id, requesting the scopes transformradar:read and, where writes are enabled, transformradar:write.
6. Sign in and allow access: The client opens a browser consent screen. Sign in with your TransformRadar account, pick the organisation if asked, and allow access. Access tokens last one hour and refresh for up to thirty days; an Org Admin can revoke a grant at any time.

How a write works: preview, then confirm

A write tool never changes the record directly. It prepares the change and returns a preview: a human-readable summary, the state before the change, a signed preview token valid for ten minutes, and a fingerprint of the current state. Only a separate commit, driven by the person confirming, applies it — and if the underlying record changed in between, the commit is refused rather than applied to stale data.

The same protocol is used by Blip inside the product, so a change proposed by an external agent and a change proposed by the in-app assistant go through exactly the same gate.

Permissions, audit and limits

A connected agent acts as the signed-in user and nothing more. It sees the projects that user can see and can prepare only the changes that user could make by hand, checked against the same permission table as the interface. Every mutation records that it was made via MCP, by whom, and when, so an audit trail can show which agent proposed a change and which person confirmed it.

Soft rate limits apply: 120 calls per minute per organisation and 60 per minute per user. A project with the Delivery workspace switched off refuses the delivery tools regardless of role.

What leaves the European chain when you connect an agent

TransformRadar’s own AI runs on Mistral in Paris and nothing leaves the EU. MCP is the one sanctioned exception: the agent you connect belongs to its own provider, and that provider handles the conversation. What leaves is what the agent asks for, on that user’s behalf, in that session — not a copy of the record. Because that is a real change in who processes the data, an Org Admin has to acknowledge it before external access can be switched on, and the acknowledgement is recorded.

Blip or MCP?

Blip is the assistant inside the product, running on the European chain, with no setup. MCP is for the AI tools your organisation already uses, on their own terms. Both are clients of the same action layer, and both end every write in a confirmation by a person. Use Blip when you are in the workspace; use MCP when you want the project record available where your team already works.

Questions people ask

Which AI clients can connect to TransformRadar?: Cursor, Claude, ChatGPT, Mistral Vibe and Google Gemini have public client ids in the catalogue. An Org Admin chooses which of them the organisation allows.
Can a connected agent change my data without me?: No. Every write tool returns a preview, and nothing is applied until the person confirms it. If the record changed in the meantime, the commit is refused.
Does using MCP send my data outside the EU?: The connecting agent’s provider handles the conversation, so what the agent asks for in a session is processed by that provider. TransformRadar’s own hosting and AI stay in the EU. An Org Admin acknowledges this before enabling external access, and it is off by default.
Do I need a developer to set it up?: No. An Org Admin enables access and allows the clients; each user then adds the endpoint and client id in their AI tool and signs in through the browser. The in-product guide shows the exact values for your organisation.
What is the difference between Blip and MCP?: Blip is the assistant inside TransformRadar, with no setup and no data leaving the European chain. MCP lets your own AI client connect from outside. Both use the same governed actions and the same preview-then-confirm protocol.
Is external MCP on by default?: No. It is off until an Org Admin acknowledges the data-egress notice and enables it, and write tiers are off until enabled separately.

Start free — no card required Back to the overview