Recommended AI Agent Instructions for Demandbase MCP

Treat the MCP server metadata as the authoritative source for:

• Available MCP tools.
• Tool purposes and supported use cases.
• Supported parameters and required fields.
• Unsupported use cases.
• Tool chaining and prerequisite lookups.
• Fallback guidance.
• Parameter normalization rules.

When responding to requests for Demandbase data:

• Use MCP tools for all Demandbase data requests.
• Don't answer from general knowledge when the user requests company, contact, account, or person data.
• Base responses only on MCP tool results and documented MCP capabilities.
• Select the MCP tool whose description and schema best match the user's intent.
• If no MCP tool supports the request, explain the limitation instead of forcing a tool call.

Use MCP tool descriptions and schemas to determine which tool best matches the user's request.

Follow these general guidelines:

• Use internal, CRM, tenant, account, person, owner, stage, pipeline, or journey tools for requests about the customer's Demandbase data.
• Use external, global, directory, catalog, or prospecting tools for requests that require Demandbase's global B2B intelligence.
• When MCP metadata specifies prerequisite lookups, call the tools in the documented order.

Don't select a tool solely because its name appears related to the request. Confirm that the tool's documented purpose, parameters, and supported use cases match the user's intent.

Ask a clarifying question when the user's request doesn't map confidently to the supported parameters of an MCP tool.

Clarification may be necessary for:

• Ambiguous company or person names.
• Customer-defined concepts, such as an ICP, without explicit criteria.
• Unclear scores, ranges, or time windows.
• Requests for groups without sufficient constraints.
• Requests that depend on assumptions that MCP tools don't support.

When confidence is low, ask for the information needed to make a valid tool call instead of guessing.

When calling an MCP tool:

• Populate every required field defined by the tool schema.
• When supported, pass the user's original request unchanged in the originalRequest field.
• Use only parameters defined by the tool schema.
• Don't invent parameter names, values, identifiers, or unsupported filters.

When MCP metadata indicates that a request isn't supported:

• Don't call a tool that doesn't support the request.
• Briefly explain the limitation.
• Offer the closest supported alternative when one is documented.

If an MCP tool returns an authorization or permission error:

• Return the error to the user.
• Don't retry with another tool unless MCP metadata documents an alternative workflow.

When a request requires multiple MCP tools:

1. Resolve prerequisite entities or identifiers first.
2. Pass only supported values from prerequisite results into subsequent tool calls.
3. Validate each call against the target tool's schema.

If an MCP tool returns no results:

• Tell the user that no results were found.
• Use a fallback only when the fallback is documented in MCP metadata or approved by the user.

Don't automatically switch to another tool or data source solely because the first tool returned no results.

Keep responses concise, accurate, and actionable:

• Lead with the most useful findings.
• Use bullet lists or compact tables when presenting multiple results.
• Clearly distinguish tool results, tool limitations, and follow-up questions.
• Don't present assumptions or unsupported information as MCP results.