Document import or processing fails
Identify the step that failed: file selection, reading, profile extraction or saving. Check whether the file opens, is password protected or exceeds size or page limits. A URL must be accessible from outside your signed-in browser.
Retry with a small file containing representative information. If that works, split the larger document by chapter or export a version with a text layer. For session errors, sign in again and check the task's status before resubmitting.
A task still processing has not failed. Avoid repeatedly submitting all documents before knowing the result; AI operations can consume credits and produce data that needs review.
Missing profile data or incorrect AI answers
First check which brand is being used. Then open the profile field and the source referenced by the answer.
- No data: add real sources or answer the missing interview questions.
- Incorrect data: identify reading errors, extraction errors or outdated information; correct the source or profile with appropriate permissions.
- Evidence does not support the conclusion: ask for an exact quotation and the passage supporting it. A reference that looks like a source is not enough.
- Old results: ask AI to reread the profile; start a new session if it retains another brand's context.
Record the question, incorrect answer, expected source and time of the issue. These are more useful than a screenshot saying “AI is wrong.”
Connection failures or lost access
Check that the URL is exactly https://mcp.antiem.io/mcp, the server is remote HTTP and authentication uses OAuth. The documentation website is not the server URL.
If tools are missing, enable Antiem in the conversation or start a new session. If tools appear but brands do not, check your account and invitations. If writing is denied, read access is insufficient: it takes both the mcp:write permission and an owner or admin role in that brand.
Authenticate again from the AI client if access was revoked or expired. When you reach call limits, stop repeated retries and read the limit notice. If connector creation is missing, check workspace permissions or try a client listed in Quick install.
Frequently asked questions
Do I need technical skills?
Not for profiles, document import or Brand Agent on the web. CLI commands are for people already using those tools; you can connect through Claude or Cursor's interface instead.
Can I use Antiem without an account?
You can read these docs without signing in. Accessing brand data through the app or MCP requires an account and permissions. Join the waitlist if you do not have beta access.
Can MCP create a brand or publish posts automatically?
MCP currently cannot create brands. Create them in the web app. Antiem does not provide automatic social publishing in these workflows.
Does a read-only prompt prevent all writes?
The prompt describes your intent; the server and token enforce actual permissions. The sign-in screen shows what the client requests, and you can only allow or deny it as a whole, so a client that asks for mcp:write can write whatever the prompt says. Even then, changes need a confirmation step and an owner or admin role. You can cut off a connection at any time in Kết nối MCP (MCP connections).
Glossary
- Brand profile: structured knowledge about identity, customers, value and expression.
- Source/evidence: material supporting a statement or conclusion that can be checked against documents.
- Boundary: something the brand explicitly will not do or become.
- Proposal: a change awaiting review, distinct from a currently active value.
- MCP: a protocol that lets AI call tools and read data from a connected service.
- OAuth: sign-in and authorization without giving AI your password.
- Scope: a connection's permission range, such as read or write.
- Credit: an allowance for AI processing, separate from MCP call counts.
MCP tool reference
AI should start with antiem_get_brands, select a brand and call antiem_get_brand before working. Match the brand name in a prompt against the permitted list.
Use antiem_get_brand_docs to find or read sources, antiem_get_brand_fields for profile fields and antiem_get_brand_gaps for missing information. Checks such as antiem_check_brand_content and antiem_check_brand_health help compare content and the profile, but do not replace human review.
Ingestion and updates require write access and their corresponding workflows. Creating a proposal differs from approving it. The table below lists the tools and permissions that exist right now; it comes from the application's registry.
| Tool | Required permission | Changes data |
|---|---|---|
antiem_get_brands | mcp:read | No |
antiem_get_brand | mcp:read | No |
antiem_get_brand_docs | mcp:read | No |
antiem_get_brand_questions | mcp:read | No |
antiem_get_brand_work | mcp:read | No |
antiem_get_brand_fields | mcp:read | No |
antiem_get_brand_gaps | mcp:read | No |
antiem_get_brand_products | mcp:read | No |
antiem_get_brand_document_status | mcp:read | No |
antiem_get_brand_documents | mcp:read | No |
antiem_get_brand_proposals | mcp:read | No |
antiem_check_brand_health | mcp:read | No |
antiem_check_brand_content | mcp:read | No |
antiem_ingest_brand_document | mcp:write | Yes |
antiem_propose_brand_update | mcp:write | With confirmation |
antiem_decide_brand_proposal | mcp:write | Yes |
On small screens, scroll the table horizontally to see every column.
Updates and support
These guides were checked against the source code on 12 October 2026. MCP tools have been renamed and grouped by task; refresh the tool list or reconnect if a client still has old names, instead of asking AI to call a tool that no longer exists.
For beta support, use the contact channel provided in your invitation or email conversation with the Antiem team. If you do not have access, start with the waitlist.
Include the client and version, time, operation, error and reproduction steps. Hide sensitive information in screenshots. Do not send tokens, passwords or full client documents when a short excerpt can explain the issue.