MCP integration
AlienMonster runs an MCP server at https://alienmonster.ai/mcp. Coding agents (Claude Code, Cursor, Claude.ai) connect to it and gain access to tools for reading and writing strategic context.
How agents connect
Two paths, depending on the client:
- Claude Code and Cursor — run
npx alienmonster connect. It fetches a long-lived bearer token and registers the server with each client. Claude Code viaclaude mcp add; Cursor via~/.cursor/mcp.json. - Claude.ai — go to Settings → MCP → Claude.ai in-app for an OAuth connector flow. See Capturing strategy in Claude.ai.
User-visible tools
Nine tools agents call during normal user workflow. Every tool requires an authenticated MCP session.
get_conversation_guide
Returns modes (explore / deep dive / quick update), interview templates, dimension state, and opening suggestions for a session. Agents should call this first when the user asks to start a session. Input: projectId, optional mode, optional template.
start_brainstorm_session
Starts an Explore-mode session. AlienMonster captures ideas, directions, chosen directions, and open questions as the conversation proceeds. Input: projectId, optional title, optional resumeSessionId. See Session types.
push_brainstorm_primitive
Captures a single piece of an Explore session — an idea, a deeper exploration, a chosen direction, an archived direction, an open question, or a connection between ideas. Input: sessionId, primitiveType, title, content, and metadata.
push_primitives
The core write path. Stores one or more structured primitives (decisions, assumptions, risks, and so on) into a project's strategy repo. Input: projectId, primitives[] with dimension, subtype, content, and optional source, links, and parent references.
get_relevant_powers
Returns the power files — target-user, business-model, market-position, tech-decisions, and risk-register — relevant to a specific coding task. Agents should call this before non-trivial edits so they don't re-litigate decisions. Input: projectId, feature_description, optional file_paths, optional include_powers. The tool matches the task against each power's inclusion rules (always, fileMatch, auto, or manual) and returns each matched power's rendered markdown content.
get_feature_context
The flagship pre-plan composition tool. Returns relevant powers, blocking assumptions, related tasks, workspace artifacts, and doctrine warnings in a single payload. Typical entry point for Claude Code planning. Input: projectId, feature_description, optional file_paths, optional keywords.
push_artifact
Saves a generated document — spec, prototype, diagram, pitch — to the strategy repo. Artifacts surface in the workspace view. Required input: projectId, filename, artifactType, title, content, category (one of: document, prototype, diagram, visualization, view). Optional: description, generatedFrom.
read_file
Reads a single file from the strategy repo. Works for all file types — core files, power files, exploration summaries, artifacts. Input: projectId, path.
list_projects
Lists the authenticated user's projects with id, name, session count, and last-updated date. Useful at session start when agents need to resolve which project to act on. No required input.
Other tools in the registry
Beyond the nine documented above, the AlienMonster MCP server registers additional tools that support facilitation flows and internal operations. They are not intended for direct use by coding agents and are not documented here:
- 11 facilitation-only tools — called by the Facilitator during sessions (brainstorm summaries, promotion of Explore primitives, doctrine edit proposals, and similar). Users experience these through conversation, not docs.
- 19 internal tools — debug, observability, chunked-upload helpers, session telemetry. Not user-facing.