Skip to content

Tools & Project Memory

ToolPurpose
get_conventionsReturns universal architecture rules, MCP toolchain usage, and communication style
list_commandsLists all available workflow commands with descriptions
list_agentsLists all available agent profiles
activate_agent("<name>")Activates a platform or cross-cutting agent, loading its full knowledge into context
list_skillsLists all available domain-knowledge skills
get_skill("<name>")Loads a specific skill’s full content
get_skill_reference("<name>", "<ref>")Loads a detailed reference document linked from a skill
search_contentFull-text keyword search across conventions, commands, and skills
suggest_nextRecommends the next command based on project command history
versionReturns the current dev-mcp server version

Project memory persists context across sessions per project. Notes, command history, and project state are tracked automatically.

Requires x-project-id header in .mcp.json — set during /setup.

ToolPurpose
get_project_contextLoad project identity, saved notes, and command history at the start of a session
save_project_note(key, value)Save a note — token decisions, component APIs, client preferences, environment details
get_project_note(key)Retrieve a saved note
list_project_notesSee all saved notes for the current project
delete_project_note(key)Delete a note from project memory
  • Agent name (saved automatically by /setup)
  • Client preferences and decisions
  • Environment details (store URL, API keys, deployment targets)
  • Design decisions that aren’t captured in CONVENTIONS.md
  • Known issues or workarounds

Every run_command call is automatically tracked per project — command name, last run timestamp, run count. This powers suggest_next recommendations.