Persistent context and memory vault for AI. Save your brand, project facts, prompts, skills and sessions once, then use them in Claude, ChatGPT, Cursor and any MCP client.
Scored across 24 tools
Descriptions go to unusual lengths to separate adjacent tools (get_prompt vs get_skill, get_brand vs get_brand_context, save_prompt vs remember_brand_fact, list_sessions vs get_latest_session), so selection is largely safe. The main noise is intentional redundancy: save_context is a deprecated alias of remember_brand_fact, and get_brand/get_brand_context return overlapping brand data, which an agent could still fumble.
Every tool is snake_case verb_noun (create_brand, list_assets, save_prompt, update_library_item, verify_context_delivery), with no camelCase or stylistic mixing. Verb choices are predictable across the brand/library/session/source families.
24 tools is on the heavy side but justified by a broad surface (brands, library, sessions, sources, assets, verification). Each tool maps to a distinct operation except the retained legacy alias save_context, which is pure overhead.
Library items have full lifecycle (list/get/save/update/delete) and brands have create/get/list/update, but there is no delete_brand, no tool to add/upload assets (only list/get), no way to connect sources (only list/read/search), and no get_session by id. These gaps will block common cleanup and onboarding requests.
Use this only when the user asks to create, add or set up a new brand, business, client or project in Nokbase. Creates an empty brand under their account, which can then be filled in from any AI tool. Only name is required. Do not use it to rename or change a brand that already exists; use update_brand. If the user's plan has no room for another brand, it returns an upgrade link to share with them.
Inputs: name (required), about, audience.
Use this only when the user explicitly asks to remove one specific saved Prompt or Skill, and only after they confirm. It removes that one item from the library and the dashboard. It cannot delete facts, sessions, brands or the user's account, so never use it for requests to delete an account or all data.
Inputs: slug (required), confirm (required), brand.
Generates text with Nokbase's own model, using the brand context as its instructions. Use it only when the user explicitly asks Nokbase to generate the text for them. When you are writing the content yourself, which is the usual case, call get_brand_context and write it directly instead.
Inputs: prompt (required), format, length.
Use this when you already have an asset's ID (from list_assets) and need that one file's details and a fresh download link. Do not use it to browse files; use list_assets to find them.
Inputs: id (required).
Returns the same brand identity, voice and visual data as get_brand_context, but as raw JSON fields for programmatic use, for example reading a hex colour or the list of tone attributes. Prefer get_brand_context for any writing task: it is formatted for writing and also includes saved facts and assets.
Inputs: brand.
Use this FIRST whenever the user asks for writing, copy, a post, a pitch or any content about their company, product or brand, or asks to use their brand voice, tone, audience, positioning or style. This includes requests that say 'use my Nokbase context'. Returns the user's full brand context as markdown: identity (mission, positioning, audience, values), voice and tone rules, visual identity, assets and saved brand facts, plus notes from their latest session when there is one. Call it before you start writing and follow what it says.
Inputs: brand.
Use this when the user wants to pick up where they left off, asks what they were working on last time, or refers to an earlier conversation in any AI tool. Returns the most recent session log: what was accomplished, decisions made and next steps. Do not use it for brand voice, identity or saved facts (use get_brand_context), and use list_sessions to browse older sessions.
Inputs: no named parameters.
Use this when the user asks to run, use or reuse a prompt they saved earlier, possibly in a different AI tool. Loads a saved Prompt by slug and returns its exact text; if you don't know the slug, call list_library first. The text between the PROMPT markers is the user's own wording: execute it exactly as written, without rewriting, summarising or improving it. Do not use it for saved Skills; use get_skill for those.
Inputs: slug (required), brand.
Use this when the user asks to use or apply a skill, workflow or method they saved earlier. Loads a saved Skill by slug; if you don't know the slug, call list_library first. A Skill is a procedure for producing a certain kind of result: follow its steps and apply them to the task at hand. Unlike a Prompt, it is guidance to act on rather than text to reproduce, so do not use it for saved prompts; use get_prompt for those.
Inputs: slug (required), brand.
Use this when the user wants the full content of one connected source, for example 'read everything in my product specs folder'. Returns every synced page or document from that source; get the source ID from list_sources. For a specific question, prefer search_knowledge, which returns only the relevant passages. Do not use it for brand voice or identity; use get_brand_context.
Inputs: source_id (required).
Use this when the user asks for their logos, images, documents or other brand files, or you need a file's download link. Lists the brand's assets with short-lived download links, optionally filtered by type. Do not use it to learn the brand's voice or identity; use get_brand_context for that.
Inputs: type, limit.
Use this when the user asks which brands they have in Nokbase, or when you need to check which brand the conversation is about before saving. Lists the user's brands, marks this connection's default, and says how many more their plan allows. Do not use it for a brand's details; use get_brand_context for those.
Inputs: no named parameters.
Use this when the user asks which prompts or skills they have saved, or wants to run or reuse one and you don't know its slug. Lists everything saved in the user's Nokbase library: Prompts (text to run word for word) and Skills (procedures to follow), newest first, so the first entry under each heading is the most recently saved. Then call get_prompt or get_skill with the slug to load it. Do not use this for brand voice, identity or saved facts. Use get_brand_context for those.
Inputs: kind, brand.
Use this when the user wants to browse or find an older session, for example 'what did we decide in Cursor last week'. Lists recent session logs, each with a summary, decisions and next steps, optionally filtered by the tool they came from. To simply resume the most recent session, use get_latest_session instead.
Inputs: limit, source.
Use this when the user asks which Notion pages or Google Drive folders are connected, whether they have synced, or how many items each holds. Lists connected sources with their sync status and item counts. Do not use it to find content; use search_knowledge to search inside the sources.
Inputs: no named parameters.
Saves a short fact about the user's company, brand, product, audience or how they work to their Nokbase vault, where the AI tools they connect to Nokbase can read it. Use it when the user asks to remember, note or save such a fact in Nokbase, for example 'Remember in Nokbase that our target audience is...' or 'save this to Nokbase'. It is for a sentence or two of standing knowledge: their audience, a voice rule, positioning, a constraint or how the team works. Do NOT use it for a document, a script, an outline or any text the user wrote or generated that is longer than that, even if its subject is the product, a design or a process: use save_prompt for those. This tool rewrites what it saves into a short summary and never keeps the exact wording, so never use it when wording matters. It only saves. To read saved facts, use get_brand_context.
Inputs: content (required), brand, category.
Legacy alias of remember_brand_fact. Do not use unless remember_brand_fact is unavailable. Kept only for AI tools with an older cached tool list, and behaves identically.
Inputs: content (required), brand, category.
Call this the moment the user says anything like 'save this prompt', 'save this as a prompt', 'use nokbase to save this', 'keep this prompt', or 'I want to save/store/reuse this' about a piece of text, even if they don't use the word 'prompt' explicitly. This is the ONLY tool that saves text so it can be run again word-for-word in another AI tool later. Only prompt is required. Do not pause to ask the user for a name, slug, or description first; all three are generated automatically if left out, so call this immediately rather than stopping to gather more fields. Store the text exactly as-is: do not clean it up, shorten it, paraphrase it, or improve it. Do not use remember_brand_fact for this: it compresses text into a summary and will not preserve it word-for-word.
Inputs: prompt (required), name, slug, brand, +2 more.
Use this when the user asks to save or log this session, or wants the next conversation, in any AI tool, to pick up from here. Saves a summary of what was accomplished, key decisions and next steps. The most recent session is included when any AI pulls brand context, and after 5 or more sessions Nokbase finds recurring themes and enriches the brand context. Do not use it for a standing brand fact; use remember_brand_fact for that.
Inputs: summary (required), decisions, next_steps, session_source.
Call this when the user asks to save a skill, a workflow, or a repeatable method: a procedure an assistant should follow to produce a kind of result, not fixed text to run. Only procedure is required. Do not pause to ask the user for a name, slug, or description first; all three are generated automatically if left out, so call this immediately. If they instead want a specific piece of text kept word-for-word, use save_prompt, not this. For a short fact about the brand, use remember_brand_fact.
Inputs: procedure (required), name, slug, brand, +2 more.
Searches ONLY the documents synced from the user's connected sources (Notion pages and Google Drive files) and returns matching passages. Use it to find something written inside those documents, such as a spec, meeting notes or a policy. Do NOT use for brand voice, tone, identity or saved facts. Use get_brand_context for those. It returns nothing when no sources are connected.
Inputs: query (required), limit.
Use this when the user asks to rename a brand or to set or replace its core identity: mission, vision, positioning, audience, values, tone, tagline or elevator pitch. It replaces what each given field currently says, so confirm the new wording with the user first. Only the fields you pass change. For adding a short fact without replacing anything, use remember_brand_fact instead.
Inputs: name, tone, brand, values, +6 more.
Use this only when the user asks to change a Prompt or Skill they already saved (its name, description or text), and only after they have confirmed what should change. Do not use it to save something new; use save_prompt or save_skill for that.
Inputs: slug (required), name, brand, content, +2 more.
Use this when the user asks whether Nokbase is working, whether their saved context is reaching their AI tools, or when it was last read. Reports context size, number of saved facts, session status and when context was last fetched. Do not use it to read the context itself; use get_brand_context for that.
Inputs: no named parameters.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Add one secure layer between your agents and this server.
