Understand and manage MCP tools

MCP (Model Context Protocol) is the interface through which Intra AI discovers and runs many integration tools. Users work with named actions in the agent builder and chat; administrators manage servers, plugins, approvals and runtime status.

Action picker for agent capabilities and integration tools

Example instance with sample data. Available controls depend on your role and configuration.

Use MCP tools#

1. Connect the required personal account under Integrations.
2. Open Tools in the agent builder and select the required actions.
3. Begin with a read-only task and state the target and boundaries precisely.
4. Review and confirm write actions when a dialog appears.
5. Open the tool group or Tool inspector in chat to inspect input, result, duration and errors.

An agent sees only approved tools. With a large catalogue, the platform can initially offer search and description tools, then load matching definitions when needed. Not every tool therefore needs to occupy the model context at once.

Status and errors in chat#

Tool calls appear in groups. The inspector shows the trace, input and result; sensitive or very large values may be shortened or omitted. Messages distinguish conditions such as timeout, rate limit, expired authentication, unreachable service, invalid parameters and withheld tools.

After a write with an uncertain outcome, do not repeat it immediately. Check the target system first to see whether it took effect.

Manager and administrator controls#

Available management views depend on role and delegated manager capabilities. Administrators can inspect server health and discovered tools, approve changed tool fingerprints, reload runtime plugins and view analytics. A manager sees only explicitly delegated areas; the audit log remains administrator-only.

New or changed tool definitions can be withheld pending approval. This prevents an unexpected description or schema from being used immediately by agents. A reload can affect current or new calls and belongs in a planned maintenance window.

Approve a tool for agents#

1. Confirm that the server or runtime is healthy.
2. Review the name, description, input and output schemas, and read/write annotation.
3. Approve a fingerprint change only when it matches the deployed plugin version.
4. Grant the tool to the intended agent.
5. Test authentication, a read case, error presentation and, where relevant, write confirmation.

Availability#

Deployment configuration defines MCP servers and runtime plugins. A configuration entry alone does not prove that the service is reachable or its tools approved. Trials do not include connectors. Some plugins, including FILERO, are available only on configured instances.

Troubleshooting#

• Server not configured: check deployment configuration and runtime address.
• Integration unavailable: check plugin health, feature switch and instance assignment.
• Authentication expired: the user reconnects the account.
• Tool withheld: an administrator reviews the fingerprint and expected plugin version.
• Runtime busy or unreachable: check health and runtime logs; do not retry blindly.
• Sub-agent write denied: delegate the tool explicitly; the instance must allow delegated writes.

Next: Integrations · Agents