Connect your AI assistantCopy Anchor Link

Connect your development assistant to 1Consent to read your app's installation settings, inspect designs, or work on customer design drafts. You need an approved 1Consent account and an app you can access. You do not need the 1Consent repository.

Server and availabilityCopy Anchor Link

Use this production remote MCP server with OAuth:

https://mcp.1consent.app/mcp

The transport is Streamable HTTP. Sign in through the browser approval flow; a customer CLI API key is not an MCP OAuth credential. The connection belongs to your assistant account, separately from the consent script installed on your website.

An older connector named “1Consent pilot” may point to a development environment. Check its actual URL. Development and production accounts, apps and approvals are separate; renaming the connector does not move it. Use a development endpoint only when your pilot instructions explicitly require it.

Choose your clientCopy Anchor Link

These instructions follow the linked client documentation, checked on 25 October 2026. They describe setup supported by each client, not completed 1Consent acceptance tests for every client or operation. Workspace policies and available controls may differ.

LovableCopy Anchor Link

Open Connectors → Add custom → MCP server, name it 1Consent, enter the production URL, keep Direct connection and OAuth, then select Add & authorize. Complete the app and permission selection below. If the MCP option is disabled, ask your workspace admin to enable custom chat connectors.

Continue with the Lovable walkthrough or Lovable's current connection instructions. Each teammate connects personally; this chat connector does not ship with the website.

CursorCopy Anchor Link

Merge this entry into your project's .cursor/mcp.json, preserving other servers:

{
  "mcpServers": {
    "oneconsent": {
      "url": "https://mcp.1consent.app/mcp"
    }
  }
}

Open Cursor's Customize MCP controls, enable the server and complete its OAuth sign-in. Keep credentials out of the JSON file. Cursor also supports user-wide configuration in ~/.cursor/mcp.json. See Cursor's MCP configuration guide.

Claude web and DesktopCopy Anchor Link

In Customize → Connectors, add a custom remote connector with the production URL and browser sign-in. Team and Enterprise members may need an owner to add it first in Organization settings → Connectors. Connect with your own account, then enable the connector in the conversation's tools menu.

Use Claude's remote connector guide. This is a hosted connector; no local server or Desktop extension is required.

Claude CodeCopy Anchor Link

In your project terminal:

claude mcp add --transport http oneconsent https://mcp.1consent.app/mcp

Open Claude Code, run /mcp, select oneconsent, and finish browser authentication. The default configuration applies to your current project. Use claude mcp list to inspect connection status. See Claude Code's remote MCP guide.

CodexCopy Anchor Link

Add the hosted server with the Codex CLI:

codex mcp add oneconsent --url https://mcp.1consent.app/mcp
codex mcp login oneconsent --scopes 1consent:context:read,1consent:config:read,offline_access

Complete the browser approval, then check codex mcp list or /mcp in the terminal UI. Local Codex clients share their host's MCP configuration. Start a new chat after setup if the existing chat does not discover the server. See OpenAI's MCP guide. offline_access requests renewable sign-in within the connection's displayed lifetime; it does not add a business capability.

For a new approval that also requests theme reads and design draft editing, sign out with codex mcp logout oneconsent, then request only the capabilities needed:

codex mcp login oneconsent --scopes 1consent:context:read,1consent:config:read,1consent:themes:read,1consent:designs:write,offline_access

Review that expanded access on the 1Consent approval page before allowing it.

ChatGPTCopy Anchor Link

Open ChatGPT Plugins, select + → Add custom MCP server, and provide a name, description, the production URL and OAuth authentication. Finish Create as a plugin, inspect the discovered tools, then install and connect it. Start a new conversation and select your plugin with @.

Your account and workspace must permit custom MCP plugins. This does not require 1Consent to appear in the public plugin directory. Follow OpenAI's connection and metadata-refresh guide if your controls differ.

Select your app and permissionsCopy Anchor Link

On the 1Consent approval page, select the organization and App scope, then the exact app you intend to work on. Similar app names are not proof of identity: confirm the returned public app ID and your website's domain/settings.

For a new context-only connection, Optional installation access offers app configuration and design-source reads, both initially off. Select the ones you need. This does not grant theme reads, analytics, or design writes. Those capabilities must be requested by the client and explicitly approved in a new authorization flow.

If a tool reports insufficient scope, follow the client's additional-access flow. If the client cannot request it, report that limitation and use the Dashboard or the customer CLI workflow for the task. Do not repeatedly retry an unchanged grant. Roles, ownership and plan/add-on requirements still apply after approval. Review the displayed lifetime; revoke unwanted connections from Connected assistants in the Dashboard.

Verify the connectionCopy Anchor Link

Run this read-only check in a chat where the connector is enabled:

Use the 1Consent connector to call get_account_context. Report the approved scope, granted capabilities, and app identity. If necessary, use list_scoped_resources to find the intended app within that scope. Do not change any configuration or drafts.

Success means an actual tool response identifies the intended account/app and access. A saved server URL or an assistant saying “connected” is not enough. If the app is missing, correct the selected account/scope before continuing.

For installation, request get_installation_context for that app with an explicit publication stage, setup strategy and loading mode. Resolve any returned blockers. For design work, use list_authoring_assets and read the selected source with read_design or read_theme, according to the permissions you approved.

Inspect configuration and evidenceCopy Anchor Link

manage_resources reads project/app metadata with separately approved resource read/write access. A context-only connection does not acquire these metadata reads. Resource write access supports project creation from an organization connection, project name/slug/description edits and app renaming using the current revision, and app creation from a covering project or organization connection using the reviewed parent revision. Current roles, activity, quotas, native provisioning and audit still apply. App grants cannot create siblings or edit their parent. Retain the same request key and exact inputs after a lost response. Required additional-app billing approval stays in the Dashboard; this tool cannot approve a charge. It does not delete resources, edit app runtime settings or publish a CMP.

read_configuration also exposes framework availability, saved environments/history, experiment candidates, deletion impact and source diffs. Use kind theme-overrides to inspect the saved app breakpoint overrides, theme setup mode and breakpoint definitions. Its configured flag distinguishes unset source from an intentionally saved empty override; basis: saved-app-settings does not describe the frozen theme in an existing published snapshot. This query does not save overrides. manage_theme reads scoped consumer impact and performs eligible clone/delete/reassignment actions; deletion requires current impact and exact replacement/consumer acknowledgments.

read_services also discovers app declarations without a collection using kind: services and groupId: null, including accepted scan findings. Collection clone/delete actions in edit_services use the exact group-impact revision. Deletion acknowledges every affected draft framework; reassignment also fences the replacement collection. Shared creation/cloning requires a connection covering the selected app's destination project or organization. These actions preserve published snapshots and do not publish the CMP.

edit_localized_content supports shared group creation when the connection covers the destination owner. Deletion uses the current group revision and acknowledges exactly every affected draft framework. A group serving consumers outside the grant cannot be deleted. A lost deletion response is recovered with the original request key and inputs; it does not require recreating the removed group.

Use read_tcf for current configuration, vendor collections, paginated GVL inventory, full taxonomy and update impact. edit_tcf_draft requires configuration write access and current TCF entitlement. Applying a GVL update needs the exact current impact revision; vendor collection changes acknowledge every affected framework. TCF runtime and publication remain subject to the canonical release restrictions.

Use manage_tcf_collection to inspect complete available active/disabled vendor memberships and scoped deletion impact, create or clone a collection, or update its name/description. Project and organization destinations require a connection covering that owner. Deletion acknowledges the exact current framework consumers; reassignment also reviews the replacement collection's revision. Hidden consumers and membership-app identities are counted without names. Use this tool's management revision, not a read_tcf revision, and retain the original request key and inputs after an uncertain response. Canonical cleanup remains available after losing the TCF entitlement; creation, metadata editing and cloning still require it.

read_scans preserves partial and inconclusive evidence. edit_scan_configuration can update recurring targets or accept/dismiss a finding; accepted findings update app service declarations. Starting a scan is not yet exposed by this increment. read_analytics returns canonical country/device/timing/framework/experiment aggregates with current plan and retention checks. Estimated events and sampled observations remain distinct; these results do not identify unique visitors.

read_audit returns scoped activity summaries with display references and encrypted one-hour cursors. Keep the same scope/date/action/outcome selection while paging. Integrity is a whole-organization check and requires an organization-scoped grant. An unanchored chain is not verified. These summaries exclude actor identities, IPs and configuration payloads.

Recover missing tools or accessCopy Anchor Link

For Lovable, inspect the connector from Connectors and reconnect there; if an old chat retains the earlier tool list, use a fresh chat. For Cursor, check the MCP server status in Customize. In Claude, check Customize → Connectors or /mcp in Claude Code. In Codex, inspect codex mcp list and re-run login for the configured server. For a custom ChatGPT MCP plugin, open its connection in Plugins, select Refresh, then start a new conversation with the plugin selected.

Do not paste OAuth tokens into chat or browser application code. You do not need 1Consent's source repository to repair the connection.

Finish a design or theme updateCopy Anchor Link

Start with get_account_capabilities to inspect the connection's approved scopes, current role, effective feature entitlements and plan quotas. Role permissions and plan features do not grant additional MCP scopes. Organization connections also receive current creation ceilings; narrower connections leave those organization limits outside their approved scope. A ceiling is not remaining capacity: each operation still checks current usage, ownership, activity and required approvals.

For a visual refresh, inspect your website's real colors, typography and spacing alongside the authorized current design/theme. Use theme tokens for styling; choose a custom design when the markup or layout must change.

The MCP can create and save eligible customer design drafts with approved write access. New designs and themes default to app ownership. Set ownerScope to project or org only when the connection covers that parent of the selected app; theme cloning supports the same destination choice. Reading an inherited asset does not grant permission to change its owner. Read the current source first, validate the complete snapshot and preserve its revision when saving. A conflict requires a fresh read and reconciliation. If a save response is lost, reread the draft before retrying; a new connection does not prove that a previous write failed.

Use manage_design to inspect customer metadata, file protections and scoped consumer/deletion impact. Its management revision covers metadata, complete own and inherited source, and current impact; use that revision for renaming, description changes, cloning or deletion. Clones retain independent source layers and file protections. Reassign bound frameworks before deleting a draft; retained published versions and migration bridges remain protected. Official designs use published reads and customer starter creation. Keep mutation keys and exact inputs after an uncertain response to recover the current owned resource or original deletion receipt.

Use list_asset_collections to discover design or TCF collections in the selected app's effective hierarchy, including before TCF setup. Follow nextCursor until it is null. Design discovery requires design read/write access; TCF discovery requires configuration read/write access.

manage_design_collection reads complete shared source, child file protections and scoped consumer/deletion impact. It creates framework-specific customer starters, edits collection metadata, clones into the same owner, and deletes eligible collections. Shared-file creation, editing, cloning and deletion use the latest collection revision and acknowledge exactly all affected design and framework IDs. The connection must cover their owners and consumers. Each affected child's complete own/inherited source is checked before a shared edit is accepted; an empty validation list means no child was checked. Source layers use the same 100-file capacity as the customer product and retain bounded UTF-8 byte limits. Bound designs and retained published versions remain protected. These operations save drafts; static validation does not prove runtime or publication readiness.

With the separate theme scope, use create_theme_draft or save_theme_draft to save customer theme tokens. Use read_configuration and edit_framework_draft with the current server revision to assign the theme/design or change appearance, behavior and eligible version pins. Content and service tools have their own scopes. Keep creation/mutation keys and exact inputs across retries; a stale edit needs a fresh read and reconciliation. The customer CLI also pushes drafts. For a local editor, Git review or Design Kit preview, ask get_1cl_reference for topic: "cli". It explains public-package setup, the separate customer API key, and the complete pull/edit/preview/push workflow. The CLI is optional when the connected tools cover the task. A hosted MCP server cannot install it; your assistant needs an authorized local terminal, or you run the steps. MCP OAuth approval does not supply a CLI API key.

runtime_preview creates a ten-minute private real CMP view for one framework. The connection owner must sign in. Source changes, expiry or revoked access stop resource delivery. Creation checks bootstrap availability; it does not prove that any browser interaction passed. OAuth tokens never belong in the preview URL or website source.

Use publication to prepare an exact single-framework Live proposal, including whether to select that framework for every visitor. Open its reviewUrl, inspect the complete changes, and approve or reject while signed in as the connection owner. Approval itself does not publish. The assistant can then execute that same releaseId; a model-supplied confirmation cannot grant approval. Changed source or expired approval requires a new proposal and review. Oversized reviews reject without approving a partial configuration; use the Dashboard workflow for those.

After a lost execution response, retry the same releaseId. Accepted publication has a durable receipt and is never repeated by that retry. Poll its status to distinguish pending, failed, converged or superseded delivery. Convergence is the server's delivery state; verify actual serving and consent behavior separately. Staging publication, promotion, rollback and explicit delivery retry are not yet exposed by this tool. Those remain in the existing Dashboard publication workflow.

These account operations require a hosted connection and fresh scope approval. The offline Git adapter provides authoring references and local validation; it does not acquire account access or publication authority.

Verify the delivered result on desktop and mobile, keyboard navigation, preferences, rejection, granular choices, reload and withdrawal with the site's actual trackers. Keep that runtime evidence separate from the MCP connection check.