MCP tool connections for AI
Discover tools from a connected MCP service and make selected operations available to an AI feature.
What it is
An MCP Tool function lets an AI feature call an operation supplied by a remote service. INSIDE discovers the tools the service offers, including their descriptions and input requirements. You select which operations to expose through AI functions.
Who uses it and why
Administrators decide which remote operations their AI features may use. Developers check the server's advertised inputs and results and configure the corresponding functions. Powerfront manages the remote connection and its authentication.
Where to find it
Ask Powerfront to set up an MCP Server connection in Settings › AI › Connections. In Settings › AI › Functions, create an MCP Tool function, select Tool Connection, then choose MCP Tool.
The Settings role with AI permission, or the Developer role, grants function and feature configuration access. Use Settings › AI › Features to allow the feature to use the function.
How it works
- INSIDE connects to the selected server and discovers its available tools.
- Selecting a tool in the function editor can populate the function's name and description. INSIDE retains the advertised input requirements for tool use.
- When an authorised AI feature calls the function, INSIDE sends the arguments to that connection and remote tool.
- The service's result returns to the AI feature. Depending on the server, it can contain text, structured data or richer content.
Automatic protocol selection tries request/response, then Streamable HTTP and legacy JSON-RPC. Compatible connections can also use older SSE or REST handling. Powerfront can select a specific mode for a server that needs it.
Configure it
Give Powerfront the server address and required authentication details. Ask Powerfront to test discovery and confirm that the intended operation is available.
Create an MCP Tool function. Choose Tool Connection and MCP Tool, review Name and Description, and leave Enabled on. Save it and add it to the intended AI feature. Use Knowledge Source if the returned information needs a recognisable source label.
Review what the remote operation does before enabling it. Tool calls can change data if the remote service provides such an operation.
Settings
| Setting | Where | Applies to | Default | Who can change it | What it does |
|---|---|---|---|---|---|
| Protocol | Settings › AI › Connections | website | Auto (recommended) | Powerfront | Chooses Auto (recommended), Request/response, Streamable HTTP or Legacy JSON-RPC for this connection. |
| Transport | Settings › AI › Connections | website | Auto | Powerfront | Chooses Auto, SSE or HTTP for compatible connections. |
| Request Mode | Settings › AI › Connections | website | Auto | Powerfront | Chooses Auto, JSON-RPC or REST for compatible connections. |
| Short Tool Prefix | Settings › AI › Connections | website | Not set | Powerfront | Adds a short prefix to distinguish tools; accepts letters, digits, underscores and hyphens. When unset, no extra prefix is added. |
| Tool Connection | Settings › AI › Functions | website | Not set | Settings role, Developer role | Selects the configured MCP connection or a compatible UCP connection for this function. A connection must be selected before tools can be discovered. |
| MCP Tool | Settings › AI › Functions | website | Not set | Settings role, Developer role | Selects one discovered operation for this function. No remote operation is selected automatically. |
| Enabled | Settings › AI › Functions | website | On for a new function | Settings role, Developer role | Makes the function available for authorised AI use on the website. |
| Knowledge Source | Settings › AI › Functions | website | None | Settings role, Developer role | Assigns a source name to information returned by this function. |
Fields
| Field | Where | Required | What it is for |
|---|---|---|---|
| Endpoint | Settings › AI › Connections | Yes | Supplies the MCP service address; an absolute address is required. |
| List Tools Path | Settings › AI › Connections | No | Sets the relative tool-list address used by REST compatibility mode. |
| Call Tool Path | Settings › AI › Connections | No | Sets the relative tool-call address used by REST compatibility mode. |
| Name | Settings › AI › Functions | Yes | Names the INSIDE function; the editor can initialise it from the selected remote tool name. |
| Description | Settings › AI › Functions | No | Explains when the AI should use this function; the editor can copy the remote tool description. |
Permissions
| Permission | Lets you |
|---|---|
| AI | With the Settings role, allows management of AI functions and feature assignments. |
| Developer | Allows management of AI functions and feature assignments. |
Limits and good to know
- Discovering a tool does not make it available to every AI feature. Function enablement and feature assignment still apply.
- Tools can be restricted by the remote credential's permissions.
- On negotiated discovery paths, INSIDE allows up to 100 pages and 2,000 tools, with a 30-second discovery deadline.
- Those paths limit an individual tool schema to one MiB, the combined catalogue to eight MiB and a response to 16 MiB. Larger results fail rather than loading without a bound.
- On these paths, tool catalogues are normally cached for five minutes. Server instructions can shorten or disable retention.
- An expired Streamable HTTP session can cause discovery to restart. Up to two session restarts are allowed while listing tools.
- Tool-name mapping keeps model-facing names within 64 characters. A prefix can help distinguish similarly named tools.
Depends on and works with
View all connections as a list
This feature uses
- AI feature settings and change historyConfigured by
The AI feature determines which tool functions it may use.
- AI functions and required inputsRequires
The selected remote operation runs through an authorised AI function.
- AI result processor templatesConfigured by
The MCP function's selected Result Processor and mode control how successful tool results are transformed.
- Knowledge sources behind AI citationsConfigured by
Knowledge Source on the MCP function supplies a recognisable source name for returned evidence.
- Shared AI service connectionsRequires
Tool discovery and calls need a configured connection.
- MCP service authenticationConfigured by
The connection's authentication choices supply request credentials where needed.
- UCP product catalogue connectionEnhances
Compatible UCP connections can supply tools through MCP.
Used by
- AI integrations and automation toolsConfigured by
The MCP function selects the remote tool and its advertised inputs.
- INSIDE AiFlow test dashboard · managed by PowerfrontEnhances
The dashboard includes MCP integration tooling.
- Knowledge sources behind AI citationsEnhances
MCP tool results can provide citation evidence.
- Shared AI service connectionsEnhances
MCP Server connections provide remote tools.
- MCP service authenticationRelated
Check MCP tool setup when authentication succeeds but the required operation remains unavailable.
- INSIDE MCP test server · managed by PowerfrontRelated
Use the fixture to investigate MCP tool discovery, execution and transport fallback.
FAQ and troubleshooting
Why is the MCP Tool list empty?
Select Tool Connection first. Ask Powerfront to check discovery, authentication and whether the server exposes tools to the configured credentials.
Why can the AI not call a discovered tool?
Check that its function is Enabled and assigned to that AI feature. The remote service must also accept the supplied arguments and authorise the operation.
Why has a newly added remote tool not appeared yet?
Discovery can use a cached catalogue. Allow its normal refresh period, or ask Powerfront to check the connection and refresh if necessary.
Why does a very large tool catalogue fail?
Negotiated discovery has tool, page and size limits. Ask the service owner to expose a smaller catalogue or smaller tool definitions.
Is an MCP tool always read-only?
No. Its behaviour belongs to the remote operation. Only enable operations that are appropriate for the AI feature and its users.