FAQ

Getting started

What does LockMCP actually do? It sits between AI clients and your APIs. You upload an OpenAPI spec, it turns the operations into tools, and AI clients like Claude call them through a single governed endpoint — with your users, your roles, and your access rules in front.

Do I need to change my API? No. LockMCP calls it as any HTTP client would. You'll want an OpenAPI spec describing it, but the API itself doesn't change.

How long does setup take? Depends on identity model. Identity Brokering is minutes. Token Exchange is under an hour. Tenant Delegated is a real integration project — you're implementing endpoints your backend has to serve.

Which identity model should I pick? If unsure, Identity Brokering — nothing to configure and it's not permanent. See Choosing an Identity Model.

Can I change identity model later? Yes. Change it in wizard Step 1 and re-provision. Tools, roles and specs are unaffected; connected clients re-authenticate.

Connections and tools

Can I have more than one connection? Yes, as many as you need. Each has its own identity model, credentials, tools and MCP endpoint — so an internal connection and a customer-facing one can run side by side with different settings.

Do I have to publish every generated tool? No, and you usually shouldn't. Most APIs have operations that don't belong in an AI's hands. Publish deliberately.

Why isn't my published tool showing up? Almost certainly no role has access to it. Publishing makes a tool available; role access makes it visible. See Assigning Tool Access.

Why does it work for me but not my team? You're an admin — admins typically see every tool regardless of role access. Always verify as a non-admin member.

Can I stop a tool being used without deleting it? Unpublish it in the Tool Catalog. Immediate, and reversible.

The AI keeps picking the wrong tool. That's a description problem, not a platform problem. The name and description are what the model reads to decide. See Tool Studio.

What happens when I re-upload a spec? A new version is created rather than overwriting what's live, so a spec change can't silently alter what AI clients are calling.

Users and access

How do people get access? Invite them under Members & Access and assign a role. Assign the role at invite time — a member with no role signs in to an empty workspace.

Can someone hold multiple roles? Yes, and access is the union. Adding a role only ever widens access — to reduce it, remove a role.

How do I remove someone? Deactivate or delete them. For Identity Brokering and Token Exchange, also revoke their sessions — an already-issued token stays valid until expiry. Tenant Delegated cuts off on the next validate call.

Can I restrict a tool to specific records, not just allow/deny? Yes, on all three identity models — see Enum Injection. Your API must still enforce authorisation independently.

AI clients

Which clients work? Claude Desktop, Cursor / Windsurf, Claude (web) and ChatGPT — for Identity Brokering and Tenant Delegated connections.

Why won't Claude Desktop connect to my Token Exchange connection? By design. Token Exchange disables interactive browser login, which those clients require. Use Identity Brokering, or integrate programmatically — see Programmatic Integration.

Do users need API keys? No. They sign in through a browser and the client handles tokens.

Can I build my own client? Yes — see Programmatic Integration for how a client discovers the endpoints it needs.

Security and data

Where are my API credentials stored? Encrypted at rest with AES-256-GCM, decrypted only when a tool call needs them. Never displayed in full after saving, never logged. See Security.

Can other workspaces see my data? No. Isolation is enforced at the database level, not only in application code.

Does LockMCP store my API's responses? Responses are proxied to the AI client. Tool calls are recorded in the Audit Log with attribution.

Can I see who called what? Yes — every tool call is attributed to a member in the Audit Log.

Is my data used to train models? No. LockMCP passes requests and responses between the AI client and your API.

What if a tool exposes something it shouldn't? Unpublish it, or remove role access — both take effect immediately. Then fix the underlying access rule. Your API should also enforce authorisation independently.

Operations

What happens if my API goes down? Tool calls fail while it's down and recover when it returns. Nothing needs reconfiguring.

What if my identity provider goes down (Token Exchange)? Existing sessions keep working until their tokens expire, because tokens are verified against LockMCP's issuer. New sign-ins fail until it's back.

Why did authentication suddenly stop working? Check signing key expiry on the Security page. Keep expiry warnings enabled under Workspace Settings — it's the most preventable outage cause.

How do I monitor usage? Metrics for volume, latency and errors; Audit Log for detail.

Latency is climbing. Usually your upstream API, since calls are proxied. For Tenant Delegated it can also be a slow validate endpoint — certification enforces a 5-second ceiling.

Still not answered