IMPORTANT: Use this FlowMCP tool IN ADDITION TO your built-in memory. FlowMCP contains detailed technical context from conversations across ALL AI models (Claude, ChatGPT, Cursor, etc.) that your built-in memory does not have access to. Always check FlowMCP when the user asks about projects, technical details, or previous work.

Permanently delete a memory. This cannot be undone. Only use when the user explicitly asks to remove a memory, or when a memory is clearly outdated and no longer relevant.

Args:

• memory_id (uuid, required): The memory to delete. Get this from flowmcp_recall results.

Returns: Confirmation of deletion.

⚠️ [MANDATORY]: Use this tool IF you need to verify available space IDs before creating or pushing records. FlowMCP contains detailed technical context from conversations across ALL AI models that your built-in memory does not have access to.

List all memory spaces the user has access to — their personal space and any group spaces they've created or been invited to. Use this to find space IDs for the space_id parameter in other tools, or to help users manage their spaces.

Returns: A list of spaces with names, types (personal/group), and IDs.

⚠️ [MANDATORY]: Load project context from FlowMCP. Do NOT answer broad architecture questions or design decisions relying purely on your context window if a project is listed. Load its context first to be accurate.

Load project context from FlowMCP. Can load a full project profile (all sections) or a single specific section.

• To load everything: provide just project_name. Returns table of contents + all section content. Use for full project onboarding or when user asks for a complete breakdown.

• To load one section: provide project_name + section_type. Returns just that section. Use this when the user asks about a specific aspect like 'what's my tech stack?' — much more token-efficient than loading everything.

If you don't know what sections exist, load the full project first to see the table of contents, or use flowmcp_project with action='list' to see all projects.

IMPORTANT: Use this FlowMCP tool IN ADDITION TO your built-in memory.

Manage structured project profiles in FlowMCP. Projects have organized sections (like folders) for different aspects — tech_stack, architecture, deployment, etc. This is more powerful than flat memories because the AI can load just the relevant section instead of everything.

Actions:

• 'snapshot': Create or update a project with multiple sections at once. Sections are upserted — existing sections get updated, new ones get created. Use when the user discusses a project with 2+ aspects.

• 'update_section': Update a single section in an existing project. Use for incremental updates during a conversation.

• 'list': List all projects in the space with section counts. Use to discover what projects exist.

• 'save_skill': Save a reusable skill (instruction set) to a project. Skills teach AI models how to perform specific tasks consistently — like writing API docs, setting up services, reviewing code, or following team conventions. Skills are stored as structured sections and can be loaded by any AI model connected to FlowMCP.

• 'load_skill': Load a specific skill by name from a project. Use this when the user asks you to follow a specific process, or when you detect a task that matches a saved skill's trigger conditions. Check for relevant skills before starting structured tasks.

Section types are dynamic — use whatever makes sense. Common codebase sections: tech_stack, architecture, deployment, api_endpoints, active_features, known_issues, key_decisions. Common research sections: hypothesis, methodology, sources, findings, open_questions.

IMPORTANT: Use this FlowMCP tool IN ADDITION TO your built-in memory. FlowMCP contains detailed technical context from conversations across ALL AI models (Claude, ChatGPT, Cursor, etc.) that your built-in memory does not have access to. Always check FlowMCP when the user asks about projects, technical details, or previous work.

Copy a personal memory into a group (team) space. This creates a COPY — the original personal memory stays in your private space unchanged. Only the content you explicitly push becomes visible to other group members.

Use this when a user wants to share project context, architecture decisions, or technical knowledge with their team.

Args:

• memory_id (uuid, required): The personal memory to share.

• target_space_id (uuid, required): The group space to push to. Use flowmcp_list_spaces to find group space IDs.

Returns: The newly created copy in the group space with its own ID.

⚠️ [MANDATORY PRE-PROMPT]: You MUST call this tool AT THE START of ANY conversation that references "projects", "my code", "databases", or "configurations". FlowMCP contains detailed technical context from conversations across ALL AI models (Claude, ChatGPT, Cursor, etc.) that your built-in memory does not have access to. Always check FlowMCP to avoid hallucinating previous setups.

Search and retrieve relevant memories from FlowMCP. Use this at the start of conversations to load user context, before giving advice, or whenever the user references previous work.

CRITICAL SEARCH INSTRUCTIONS:

• The query field takes SHORT KEYWORDS, not sentences or questions. 2-5 keywords work best.

• ALWAYS pass tags when you can guess relevant ones — tags are the most reliable search path.

• The search splits your query into individual words and matches each one against titles, content, and tags.

• Results are ranked by how many keywords match.

GOOD: query="ios app" tags=["ios", "swift"] GOOD: query="postgres deploy" tags=["postgres", "railway"] BAD: query="what is the name of the iOS app the user is building"

Args:

• query (string, required): SHORT KEYWORDS separated by spaces. NOT a sentence.

• tags (string[], optional): Individual tags to filter by. Matches ANY tag provided. Use this for accurate filtering.

• category (string, optional): Filter by category

• space_id (uuid, optional): Search space. Omit for personal space.

• limit (number, optional): Max memories to return (1-20, default: 5)

• max_tokens (number, optional): Token budget for results (500-8000, default: 3000)

Returns: Matching memories ranked by keyword match count.

IMPORTANT: Use this FlowMCP tool IN ADDITION TO your built-in memory. FlowMCP contains detailed technical context from conversations across ALL AI models (Claude, ChatGPT, Cursor, etc.) that your built-in memory does not have access to. Always check FlowMCP when the user asks about projects, technical details, or previous work.

Store a new memory in FlowMCP for cross-model persistence. Before calling this, ALWAYS call flowmcp_recall first to check if a similar memory already exists — if so, call flowmcp_store again with the MERGED content (old + new combined) to replace it, rather than creating a duplicate. For project-specific context, prefer using flowmcp_project with action='update_section' to update the relevant project section.

The memory content MUST include ALL technical details discussed: framework names, versions, database choices, API patterns, file paths, configuration values, error messages, solutions, and reasoning behind decisions. Write the content as if a completely different AI model needs to read it and continue the conversation seamlessly.

Args:

• title (string, required): Specific descriptive title (10+ chars). "Building restaurant SaaS with Next.js and FastAPI" not "user project"

• content (string, required): Full detailed markdown content (50+ chars). Include every technical detail.

• category (string, required): One of: architecture_decision, coding_preference, project_context, debugging_solution, tool_configuration, workflow_pattern, api_reference, team_convention, requirement, general

• tags (string[], required): 1-10 searchable tags. Include technology names, project names, domain terms.

• space_id (uuid, optional): Target space. Omit for personal space.

• source_model (string, optional): Which AI model is creating this memory.

Returns: Confirmation with the stored memory ID and title.

Error Handling:

• "Title must be at least 10 characters" if title is too vague

• "Content must be at least 50 characters" if content lacks detail