Privacy and permissions

Every MCP call runs as you. This page explains what that means for what the assistant can see, how each request is secured, and how to cut a client off.

Your identity, your scope

MCP calls run under your valueIQ identity. The assistant sees exactly what you see in the app: nothing from other accounts, and nothing outside your workspace scope. If a deal is in a workspace you are not a member of, the assistant cannot read it. If your role in a workspace is Viewer, the assistant cannot make edits there either. The four roles are Owner, Admin, Member and Viewer, in that order of access.

What reads, what writes and what costs credits

Read only, free: listing and opening deals, products, value stories and their version history, documents, comments, invites, coaching threads, market values, a product's edit history, comparable-company proposals, DIY Reality Checks, pricing analysis, CRM scorecards, workspace summaries and the credit balance. Coaching, objection responses and follow-up drafts are also free.
Writes, free: updating deals, customer variables, deal notes and deal context, driver applicability and driver notes, deal members, CRM links, story top drivers, presentation locks, story wording, currency and box visibility, story versions and bookmarks, archiving and restoring stories, comments and invites, workspace invitations (which email the invitee), documents, workspace settings, archiving and restoring deals, moving a product through its edit history, accepting or rejecting a comparable-company proposal, sign-off and session checkpoints.
Writes that draw credits: creating a product or a product variant, creating a deal, restoring an archived product (and each deal you bring back with it), running a DIY Reality Check, and exporting a deck as PPTX or PDF. Rates are listed in the MCP Server overview.

A comment you post on a story is visible to everyone with access to that story, including customer partners.

If every workspace you belong to gives you read-only access (the Viewer role), the assistant is not offered any write action at all. Editing a product's structure (rewording or restructuring drivers, moving a product through its edit history, creating products and variants) also needs the value model edit permission. Owners and Admins have it; a Member needs an Owner or Admin to grant it. Renaming, archiving and restoring a product, and changing workspace settings, need an Owner or Admin. Editing a value story, estimating from comparables and running a DIY Reality Check need to be a member of the deal's workspace. Only a story's owner can bookmark one of its versions.

Archiving a product also archives its variants and every deal on them, so the assistant never does it in one step: it lists what would be archived and archives only after you confirm that list. It never archives a product as a side effect of another request.

Documents you attach

A file you attach through the assistant goes through the same checks as an upload in the app and is stored against your own profile on the deal. Only members who can see the deal can see the file. Only the person who uploaded a document can remove it from the deal's chat.

Row-level security

valueIQ enforces row-level security on every database table. Each request mints a short-lived token for you. Reads and most edits run under your own database permissions, so they cannot return rows you would not be able to see in the app. A few actions, such as rebuilding a model, run on the server after valueIQ checks your role and workspace membership. Every call is logged.

Authentication is OAuth 2.1 with PKCE. You sign in through your browser and approve the connection once per client. There are no static, long-lived credentials: no API key to store in a config file and no token to copy. Tokens are minted per request on the valueIQ side.

Your deal data is not used for training

valueIQ does not use deal data passed through MCP connections for model training.

Revoking a client

Go to Workspace Settings → Developer and remove the connection. The client loses access immediately. To reconnect later, use the client's setup steps again and approve the sign-in when prompted.

Input validation and prompt injection

All inputs are validated before a tool executes. Content you pass into a coaching or drafting request, such as pasted notes or an objection in the buyer's words, is wrapped in explicit delimiters before it reaches the model, so caller-supplied text cannot be read as an instruction. This closes a prompt-injection path where pasted content could otherwise have been treated as a continuation of the instructions.

Audit log

Every MCP call is logged. Enterprise customers can access the full audit log through the API.