MCP tools

Inth MCP provides one tool per REST operation. Tools return the same resource shapes as the REST API. The client should ask for confirmation before it calls a tool that changes data; every tool carries annotations that say whether it is read-only or destructive.

Organization tools

list_organizations

Lists the Inth organizations that your signed-in account can access, including your role in each organization. The token must include organizations.read.

This tool has no arguments.

Example prompt:

List my Inth organizations and show my role in each one.

get_organization

Gets one organization. The token must include organizations.read.

Every other tool that takes an organizationId treats it as optional: without it the saved active organization of your account is used. If that choice is unavailable and you belong to multiple organizations, provide the ID explicitly. Tools that take a member, invitation, project, scan, or item ID find the organization from that ID.

create_organization

Creates an organization owned by your signed-in account and sets up its billing on the free plan. The token must include organizations.write. Without it the call returns a tool error whose code is INSUFFICIENT_SCOPE.

Example prompt:

Create an Inth organization named Acme with the slug acme.

You can own up to 50 organizations. A slug that is already taken is refused.

Member tools

list_members

Lists the members of an organization with their roles. The token must include members.read. Any member can read the roster.

Example prompt:

Who is in my Acme organization, and what role does each person have?

update_member

Changes a member's role. The token must include members.write, and you must be an admin or owner. Only owners can grant or take the owner role, you cannot change your own role, and the last owner cannot be demoted.

remove_member

Removes a member from an organization. The token must include members.write, and you must be an admin or owner. Only owners can remove owners, you cannot remove yourself, and the last owner cannot be removed.

list_invitations

Lists pending invitations. The token must include members.read, and you must be an admin or owner.

invite_member

Emails an invitation to join an organization. The invitation expires after 48 hours. The token must include members.write, and you must be an admin or owner; only owners can invite owners. An address that is already a member or already invited is refused.

Example prompt:

Invite kim@acme.example to my Acme organization as an admin.

cancel_invitation

Cancels a pending invitation. The token must include members.write, and you must be an admin or owner.

If acceptance or deletion wins a race with cancellation, the call returns CONFLICT. Refresh the member list before trying again.

Project tools

list_regions

Lists the regions available when creating a project. Use the returned region ID with create_project.

This tool has no arguments.

Example prompt:

Which regions can I use for a new Inth project?

list_projects

Lists projects with their consent settings. You can list every project you can access or limit the result to one organization. The token must include projects.read.

Example prompt:

List the projects in my Acme organization.

get_project

Gets one project with its consent settings. The token must include projects.read.

create_project

Creates a project and its consent runtime. Admins and owners can use this tool when the organization has API access and the token includes projects.write.

Example prompt:

Create an Inth project named Docs in the Frankfurt region for the Acme organization. Trust https://docs.example.com and show Inth branding.

Check the organization, region, project name, and trusted origins before you approve the call. To find the required IDs first, ask the client to run list_organizations and list_regions.

update_project

Renames a project, changes its description, or changes its consent settings. Arguments left out are unchanged. Renaming a project also changes its slug and dashboard URL. The token must include projects.write, and you must be an admin or owner. Removing branding needs a plan that includes it.

Example prompt:

Add https://staging.example.com to the trusted origins of the Docs project.

delete_project

Archives a project and its consent runtime. The token must include projects.write, and you must be an admin or owner. Confirm before approving.

API key tools

list_api_keys

Lists the organization API keys. Secrets are never returned. The token must include api-keys.read, and you must be an admin or owner.

create_api_key

Creates an organization API key with the rate limit of the current plan. The secret is returned once. The token must include api-keys.write, and you must be an admin or owner.

Example prompt:

Create an API key named "CI deploy" for my Acme organization.

roll_api_key

Replaces the secret of a key. The old secret stops working immediately and the new one is returned once. The token must include api-keys.write, and you must be an admin or owner.

delete_api_key

Revokes a key. The token must include api-keys.write, and you must be an admin or owner.

Code Audit tools

List tools take an optional organizationId; without it the active organization of your account is used. Tools that take a scan, repository, or item ID find the organization from that ID.

list_code_audit_repositories

Lists the GitHub repositories the Inth App can reach, with the projects each is linked to, ordered by repository ID. The token must include code-audit.read.

Returns items, nextCursor, and hasMore. Each request reads only the requested repository page and its project links.

list_code_audit_scans

Lists scans, newest first, with status, finding counts, and whether the report is a free preview, locked, or fully unlocked. The token must include code-audit.read.

Returns items, nextCursor, and hasMore. Pass nextCursor as cursor to read older scans.

get_code_audit_scan

Gets one scan with its progress. The token must include code-audit.read.

start_code_audit_scan

Starts a scan of the production branch of a connected repository. The first scan of a repository is a free preview; the previous preview must be unlocked before another scan starts. The token must include code-audit.write, and you must be an admin or owner. Confirm before approving: a scan spends credits when its report is unlocked.

Example prompt:

Scan the acme/website repository with Inth Code Audit and tell me what it found.

get_code_audit_scan_request

When start_code_audit_scan returns status: "starting", poll this tool with its preparationId and repositoryId until it returns the scan. Requires code-audit.read or code-audit.write.

get_code_audit_scan_issues

Reads the findings of a scan. A free preview returns the visible findings plus lockedCount and unlockCredits. The token must include code-audit.read.

unlock_code_audit_scan

Spends credits to unlock the full report of a free preview. The token must include code-audit.write, and you must be an owner. Confirm the credit cost before approving; the call fails with INSUFFICIENT_CREDITS when the balance is too low.

Inbox tools

list_inbox_items

Lists findings in the Inbox, most recently updated first. The token must include inbox.read.

Example prompt:

What open P0 and P1 findings are in my Acme Inbox?

get_inbox_item

Gets one finding with its code evidence. The token must include inbox.read.

A runtime finding, raised when a scan sees a vendor on your site, also has runtimeVendor. It holds the project ID, the vendor ID, the hosts its evidence names, and evidence of how the vendor loaded: the page, the script URL, and the Google Tag Manager container when Tag Manager loaded it. The hosts include cookie domains, so a host on the list was not always contacted. Other findings have runtimeVendor: null.

update_inbox_item

Sets the status of a finding. Pass the version from the item as last read; a stale version is refused so two people cannot overwrite each other. The token must include inbox.write, and you must be an admin or owner.

Example prompt:

I fixed the logging finding in the checkout handler. Mark it resolved.

create_inbox_github_issue

Opens a GitHub issue in the connected repository for a code finding. Returns the existing issue if one was already created. The token must include inbox.write, and you must be an admin or owner.

Vendor tools

list_project_vendors

Lists the vendors seen on a project, most recently seen first, with each vendor's consent category, state, and review state. A vendor in state active has no decision yet. The token must include projects.read.

Example prompt:

Which vendors on the marketing site still need a decision?

get_project_vendor

Gets one vendor seen on a project: the hosts its evidence names, and where and how the scan saw it load, including the script URL and the Google Tag Manager container when Tag Manager loaded it. inboxItemId is the vendor's runtime finding in the Inbox, in any status, or null. The token must include projects.read.

decide_project_vendor

Approves or ignores a vendor seen on a project. Approving accepts the data sharing. Either decision closes the vendor's runtime finding in the Inbox, and later scans keep the decision. Take projectId and vendorId from list_project_vendors or from runtimeVendor on get_inbox_item. The token must include vendors.write, and you must be an admin or owner.

Example prompt:

The Hotjar finding on the marketing site is expected. Approve it.

Billing tools

get_billing

Reads the plan, tier, credit balance, and automatic top-up settings of an organization. Read only; purchases stay in the dashboard. The token must include billing.read.

Example prompt:

How many Inth credits does Acme have left, and is auto top-up on?

An uncertain issue creation or failed persistence retains a reservation, so retries return CONFLICT until the result is reconciled. Repository and installation-token checks run before reservation, so preparation failures remain retryable. Refresh the item and contact support if the issue link does not appear.

Agent tools

The Inth Agent answers consent, privacy, legal research, and c15t questions. A chat is private to you. A reply can take minutes, so starting a chat or sending a message returns once the agent has accepted it, with status running. Call get_agent_chat until status is idle to read the reply, or open dashboardUrl to watch it in the Dashboard. Replies use your organization's credits; without enough, the send is refused with INSUFFICIENT_CREDITS. A token without agents.read gets messages and title empty. Reusing a clientMessageId with different text or another Inbox item returns CONFLICT, and so does reusing one whose turn stopped before the agent started it. Organization API keys cannot use these tools.

start_agent_chat

Starts a chat with a first message. If the response is lost, call it again with the same clientMessageId: you get the chat the first call started, and the agent does not answer twice. After an error, use a new ID. The token must include agents.write.

Example prompt:

Ask the Inth Agent whether our German site needs consent before it loads analytics.

get_agent_chat

Gets a chat with its messages and status. A reply with failed: true ended without an answer. So did a turn whose last message is still yours once status is idle; send it again. The token must include agents.read.

send_agent_message

Sends a follow-up message to one of your chats. A chat answers one message at a time: a send while status is running returns CONFLICT, unless it repeats the running message's clientMessageId. The token must include agents.write.

Example prompt:

Ask the agent the same question for France.

Feedback

submit_feedback

Sends a report to the existing Inth feedback inbox and returns reference and alreadySubmitted. Requires the usual MCP OAuth authentication and mcp:tools, with no additional product scope. It does not charge Credits.

Required arguments are category, one of bug, friction, docs_mismatch, feature_request, or performance; and message, a description of the problem.

Optional arguments are surface, one of api, mcp, cli, docs, sdk, dashboard, or other; and expected, actual, reproduction, requestId, operation, clientName, and clientVersion.

Report unexpected failures, misleading tool descriptions, missing capabilities, or workarounds. Follow the user's permission to send feedback. Include expected and actual behavior, without credentials, personal data, customer payloads, or full transcripts. Submit once per distinct issue; do not report every routine validation error. If reporting fails, continue the original task.

The quota is ten new reports per user per UTC clock hour. Sending the same report again within the same UTC day returns the existing reference with alreadySubmitted: true and does not use quota. An operator can pause intake without hiding saved feedback.