MCP Prompts Overview
Phoenix exposes MCP prompts (workflows) that guide AI assistants through structured, repeatable research tasks. A prompt combines a templated message body with a small set of named arguments. Clients discover prompts via prompts/list and render them via prompts/get.
What are MCP Prompts?
MCP prompts are pre-built workflows that:
- Chain multiple tool calls together
- Provide structured templates for research
- Ensure consistent output formatting
- Reduce complexity for common use cases
Available Prompts
The previous built-in prompts (tam_analysis, pov_trigger_analysis, company_search_guide, research_rule_generator) have been retired. Phoenix is rolling out a new set of workflows under the GTM marketplace; this overview will list them as they ship. In the meantime, prompts/list returns an empty array — your MCP client will simply show no prompts. Tools (/mcp-tools/overview) remain fully functional.
Prompts vs Tools
When to use Prompts
- Complex multi-step research workflows
- Repeatable processes that combine multiple tools
- Guidance on how to structure queries
- Template generation
When to use Tools directly
- Simple single-step queries
- Custom workflows not covered by prompts
- Real-time data retrieval
- Exploratory research
Using Prompts with MCP Clients
When prompts are available, MCP clients render them in a UI affordance (a slash menu or prompt picker). Invocation goes through the standard MCP prompts/list and prompts/get flow — no Phoenix-specific protocol.
Prompt Output Format
Prompts return a single user-role message containing the rendered template, with caller-supplied arguments substituted into {{varName}} placeholders. The rendered output is bounded; an oversized rendering (≥ 100 KB) returns an error rather than a giant message.
Coming Soon
A new generation of prompts is in development. Expect to see them surface in prompts/list in the coming weeks.