kb_attachments_archive
ChatGPTSoft-archive an attachment (sets archived_at). Reversible via kb_attachments_restore.
kb_attachments_begin_upload
ChatGPTBegin a short-lived direct attachment upload for a client that can transfer local file bytes. Provide document_id and the original filename; optionally provide expected_sha256. The result contains an HTTPS upload_url ending in #token=<capability>. The fragment is local to the client and is never sent to the server: extract its token and POST the raw file bytes with Content-Type: application/octet-stream and Authorization: Bearer <token>. Send no multipart envelope; upload_url is the endpoint, not a file field, not the fragment. The capability expires in five minutes and succeeds once. The returned size limit is authoritative. Never put the token in a path, query, tool argument, or log.
kb_attachments_get
ChatGPTRetrieve attachment metadata, including the attachment_uuid used to embed the file in a document body as kb-attachment://<uuid>. Pass include_download_url=true to also mint a short-lived download URL.
kb_attachments_import
ChatGPTSave the actual bytes of a user-provided file reference as a downloadable document attachment. Pass the host-provided file object; the server downloads its public HTTPS URL with a bounded transfer and validates the bytes. Never invent a URL or substitute a description for the file. If the chat host cannot provide a downloadable reference, say so; a client holding local bytes can use kb_attachments_begin_upload. Optional expected_sha256 checks integrity without returning a hash. Returns attachment_uuid; embed  for an image or [label](kb-attachment://<uuid>) for a file. Active downloadable attachments remain until explicitly archived. Temporary source URLs and provider file IDs are not retained.
kb_attachments_import_url
ChatGPTAttach the exact original bytes from a public HTTPS source URL to an existing document. Pass source_url as an ordinary string and the mandatory expected_sha256 independently supplied by the source service; never invent a URL, checksum or substitute file. Optional expected_size_bytes must match exactly. The server validates each redirect, bounds the transfer and verifies integrity before storage. Returns attachment_uuid without changing the document body or revision. To link the file, separately edit the document using [label](kb-attachment://<uuid>). Active downloadable attachments remain until explicitly archived. Source URLs and checksums are not returned or retained as source metadata. A failed one-use source must be replaced with a fresh source capability before a deliberate retry; the server never replays a failed request. For a chat-host file object use kb_attachments_import; for local bytes use kb_attachments_begin_upload.
kb_attachments_list
ChatGPTPaginated list of a document's attachments. Active (non-archived) only by default. Each row carries the attachment_uuid used to embed the file in a document body as kb-attachment://<uuid>.
kb_attachments_restore
ChatGPTRestore a soft-archived attachment (clears archived_at). Reverse of kb_attachments_archive.
kb_attachments_upload
ChatGPTUpload a small file you already hold as text or bytes, up to 64 KiB. Supply exactly one of content_text or content_base64. Never invent file bytes or write out base64 for a photo or a file shown in chat. Use kb_attachments_import for a real downloadable file reference, or kb_attachments_begin_upload for a client that can transfer local bytes directly. When neither is available, give the user the document permalink, ask them to open it and use Attach, then call kb_attachments_list. The web file limit is deployment-configured; an inline refusal states the live limit. Returns attachment_uuid: embed as  or [label](kb-attachment://<uuid>). Unreferenced inline uploads are eventually swept; active downloadable uploads remain until explicitly archived.
kb_attachments_view_image
ChatGPTPresent an existing Brainattic image attachment. Native MCP image content is the default; presentation_mode=auto falls back to a short-lived Brainattic presentation URL only when the normalized rendition is above the 1 MiB native envelope. Use native to require ImageContent, or url to explicitly request the Brainattic presentation capability. This tool never exposes MinIO or returns the original image bytes as a download.
kb_chunks_retrieve
ChatGPTRetrieves relevant passages from the connected private knowledge base. Call this whenever a question depends on the organisation's stored knowledge; the model may not otherwise have access to it. Returns a flat list of passages ordered by relevance; pass them as context when generating your answer and cite the source document for each claim. For "show me documents about X" browse-style queries, prefer kb_documents_search.
kb_documents_archive
ChatGPTSoft-archive a document (sets archived_at). Reversible via kb_documents_restore.
kb_documents_create
ChatGPTCreate a new document. slug is auto-generated from title when omitted. project_slug must exist; group_slugs are created on demand.
kb_documents_delete
ChatGPTPermanently delete a document and all its revisions. Admin-only. Irreversible — archive first if in doubt.
kb_documents_edit_apply
ChatGPTApply an ordered batch of edits to an open edit session, all-or-nothing — if any edit cannot be applied (no match / ambiguous / overlapping / out of range) the whole batch is rejected and the buffer is left unchanged. Edit types: "replace" (old_string unique unless replace_all=true, new_string), "line_range" (start_line/end_line/content), "section" (heading + mode replace_body|insert_after_heading|delete_section, with level/parent to disambiguate). Returns the cumulative base→working unified diff; pass an empty edits array to just re-fetch the diff without mutating.
kb_documents_edit_begin
ChatGPTOpen a surgical edit session on a document and snapshot its current revision. Stage edits with kb_documents_edit_apply, then kb_documents_edit_commit (lands one revision) or kb_documents_edit_discard. Calling this again for a document you already have an open session on resumes that session (resumed=true).
kb_documents_edit_commit
ChatGPTCommit an open edit session: flush its working buffer to a single new revision (title / project / groups preserved). Honors the begin-time revision snapshot — a concurrent edit returns kb.documents.stale_revision and leaves the session open to rebase or discard. Returns no_changes=true (no revision written) when the buffer equals the current document body.
kb_documents_edit_discard
ChatGPTDiscard an edit session and throw its working buffer away — no revision is written. Idempotent: discarding an already-closed session is a no-op.
kb_documents_get
ChatGPTRetrieve a single document with its full markdown body. Optional include array bundles related rows in one round-trip: revisions / attachments / groups / project. Use kb_documents_get_meta when the body is not needed.
kb_documents_get_by_slug
ChatGPTCheap slug-based lookup returning minimal metadata (no body). Returns null when no document with that slug exists in the tenant. Use kb_documents_get for the full body.
kb_documents_get_meta
ChatGPTCheap id-based lookup returning minimal metadata (no body). Useful for "is this doc tagged right?" follow-ups after a search or list. Use kb_documents_get when the body is needed.
kb_documents_list
ChatGPTPaginated browse of documents in the tenant. Filters: project_slug, group_slugs, archived, include_hidden. Sort: updated_desc (default) | created_desc | title_asc. By default, docs belonging to hidden projects (kb_projects_hide_toggle) are excluded — matches the web /docs default view. Pass include_hidden=true to include them. Hidden does NOT affect kb_documents_search or kb_chunks_retrieve. When project_slug is explicit, the caller named the project so the hidden filter is irrelevant and is bypassed.
kb_documents_restore
ChatGPTRestore a soft-archived document (clears archived_at). Reverse of kb_documents_archive.
kb_documents_search
ChatGPTHybrid keyword+vector search across documents in the caller's tenant. Returns ranked results with snippets. mode: hybrid (default) | keyword | vector. For passage-level retrieval to ground a generated answer, prefer kb_chunks_retrieve; this tool returns doc-grouped results shaped for browse and search UIs.
kb_documents_update
ChatGPTSave a new revision of an existing document. Omit title/body_md/project_slug/group_slugs to preserve existing values. Pass expected_revision_id for optimistic-concurrency protection; omit for last-write-wins. Returns no_changes=true (and inserts no revision) when the save would change nothing.
kb_groups_archive
ChatGPTSoft-archive a group (sets archived_at). Reversible via kb_groups_restore. Archiving does NOT touch the group's document links — they are preserved; the group just leaves active listings and the sidebar.
kb_groups_attach_document
ChatGPTAdd a document to a group. Idempotent — no-op if the document is already in the group.
kb_groups_create
ChatGPTCreate a group. Idempotent on slug — returns the existing group if the slug already exists.
kb_groups_delete
ChatGPTDelete a group. Documents tagged with it lose this group tag.
kb_groups_detach_document
ChatGPTRemove a document from a group. Idempotent — no-op if the document is not in the group.
kb_groups_get
ChatGPTRetrieve a single group by slug (active or archived). The archived_at field is null when active, an ISO 8601 timestamp when archived.
kb_groups_list
ChatGPTPaginated browse of groups in the tenant, ordered by name. Active (non-archived) groups only by default; pass archived=true to list archived groups instead. Each row carries archived_at. Returns total (the full match count, independent of limit/offset) alongside the page.
kb_groups_pin_toggle
ChatGPTToggle a group's sidebar-pin state. Pass pinned=true to pin, false to unpin.
kb_groups_restore
ChatGPTRestore (un-archive) a group: clears archived_at so the group returns to active listings and the sidebar. Reverses kb_groups_archive.
kb_groups_update
ChatGPTRename or update a group's name, slug, or description.
kb_health_ping
ChatGPTHealth check; answers ok and your default workspace slug. Call kb_tenants_list to see every workspace the token grants.
kb_projects_archive
ChatGPTSoft-archive a project (sets archived_at). Reversible via kb_projects_restore. Archiving does NOT touch the project's documents — they keep their project link and stay reachable; the project just leaves active listings, the sidebar, and the dashboard "Active projects" count.
kb_projects_create
ChatGPTCreate a new project. slug is auto-generated from name when omitted.
kb_projects_delete
ChatGPTDelete a project. Documents assigned to it become project-less (project_id set to null).
kb_projects_get
ChatGPTRetrieve a single project by slug (active or archived). The archived_at field is null when active, an ISO 8601 timestamp when archived.
kb_projects_hide_toggle
ChatGPTToggle a project's hidden flag. Hidden projects are filtered out of standard browse listings (the /docs page, the /admin/projects page, the sidebar Projects collapsible, the document editor's project picker) until "Show hidden" is selected. Hidden does NOT affect search or agent retrieval — kb_documents_search and kb_chunks_retrieve still surface content from hidden projects so grounding stays complete. Pass hidden=true to hide, false to unhide.
kb_projects_list
ChatGPTPaginated browse of projects in the tenant. Active (non-archived) projects only by default; pass archived=true to list archived projects instead. Hidden projects (see kb_projects_hide_toggle) are excluded by default; pass include_hidden=true to include them. Each row carries hidden + archived_at so callers can branch. Returns total (the full match count, independent of limit/offset) alongside the page.
kb_projects_pin_toggle
ChatGPTToggle a project's sidebar-pin state. Pass pinned=true to pin, false to unpin.
kb_projects_restore
ChatGPTRestore (un-archive) a project: clears archived_at so the project returns to active listings, the sidebar, and the "Active projects" dashboard count. Reverses kb_projects_archive.
kb_projects_update
ChatGPTRename or update a project's name, slug, or description.
kb_reminders_cancel
ChatGPTCancel a single reminder (by reminder_id) or an entire schedule (by schedule_id). Idempotent — already-cancelled rows are no-ops. Only reminders and schedules you created or receive are reachable; anything else answers not_found.
kb_reminders_create
ChatGPTCreate a one-off reminder for one or more tenant members. When it fires, it SENDS a message through email, Slack, or SMS, and a message that has been delivered cannot be recalled — cancelling only stops a reminder that has not fired yet. Create one only when the user asked for it, addressed to the people the user named. Omit recipients to send to yourself; omit channels to use each recipient's preferred channel.
kb_reminders_create_schedule
ChatGPTCreate a recurring reminder schedule. Validates the RFC 5545 rrule before persistence; expansion to per-occurrence kb_reminders rows happens in the materialiser. Every occurrence SENDS a message through email, Slack, or SMS to the selected recipients, and a delivered message cannot be recalled — cancelling the schedule only stops occurrences that have not fired yet. Create one only when the user asked for a repeating reminder, addressed to the people the user named.
kb_reminders_get
ChatGPTFetch a single reminder by id with its delivery rows. Only reminders you created or receive are reachable; anything else answers not_found.
kb_reminders_list
ChatGPTPaginated browse of the actor's reminders on the active tenant, most-due-soon first. Filters: status, due window. Returns total (the full match count, independent of limit/offset) alongside the page.
kb_revisions_get
ChatGPTRetrieve a specific document revision with its full markdown body.
kb_revisions_list
ChatGPTList the revision history of a document. Newest first.
kb_revisions_restore
ChatGPTPromote a past revision to a new current revision. The body is copied forward — historical rows are never mutated (ADR-0027).
kb_rules_get
ChatGPTFetch the rule document linked to one scope. Pass scope=tenant|project|group; for project/group also pass owner_slug. Returns linked=false with rule=null when the scope has no rule doc. Use kb_rules_list_for_context instead when you want every rule that applies to an authoring context at once. Rules are advisory conventions, not machine-enforced.
kb_rules_list_for_context
ChatGPTLists the rule documents that apply to a given authoring context — the conventions your authoring should follow. Returns a list of MCP resource URIs in canonical order: tenant rules first, then project rules, then group rules (one per group, in the order you supplied). Each entry carries the rule doc body inline by default (body_md) so one call gives you everything you need — no separate resources/read fetch required. Pass include_bodies=false to get URIs only. Empty list means no rules apply at this scope. Rules are advisory, not machine-enforced; agents are expected to read and apply them.
kb_rules_set
ChatGPTLink an existing document as the rule doc for a scope. Pass scope=tenant|project|group, document_id, and (for project/group) owner_slug. The document is promoted to the matching rules kind and the scope's rule link is set to it. Idempotent. The document must live in the same tenant. Tenant-scope changes require admin; project/group require editor. Use kb_documents_update to edit the rule body.
kb_rules_unset
ChatGPTClear the rule link for a scope (sets the owner's rule FK to null). Pass scope=tenant|project|group and (for project/group) owner_slug. The rule document itself is not deleted — only the link is removed. Idempotent. Tenant-scope changes require admin; project/group require editor.
kb_skills_archive
ChatGPTArchive a skill and expire its open edit sessions. Archived skills leave live discovery; reports whether its state changed.
kb_skills_create
ChatGPTOpen a draft create session. Stage the skill body, header and supporting files, then commit it. Returns the session handle and expiry.
kb_skills_delete
ChatGPTQueue permanent deletion of a skill and its history and supporting files. Returns an operation handle; poll kb_skills_deletion_status for the outcome.
kb_skills_deletion_status
ChatGPTRead the private transient state of your own queued skill deletion or purge operation.
kb_skills_edit_apply
ChatGPTStage one header patch, a complete Markdown body, or an atomic batch of body edits in your skill edit session. Set operation to set_header, set_body or apply_body_edits and supply only its matching fields, body_md or edits payload. Header fields are name, description, license, compatibility, metadata and allowed-tools; allowed-tools is a grant request. Body edits use replace, line_range or section. Returns the session handle, expiry and fixed warnings.
kb_skills_edit_begin
ChatGPTOpen or resume your live edit session on a skill. Returns the session handle and expiry.
kb_skills_edit_commit
ChatGPTCommit your edit session as one immutable skill revision. A stale session remains open for rebase; an unchanged bundle closes with no_changes=true.
kb_skills_edit_discard
ChatGPTDiscard your edit session without publishing its staged changes. Reports whether a live draft was discarded.
kb_skills_edit_move_file
ChatGPTStage a supporting-file move in your edit session, preserving its bytes. Commit the session to publish the move.
kb_skills_edit_put_file
ChatGPTStage or replace a supporting file in your edit session. Accepts UTF-8 text or base64; decoded bytes are capped at 1 MiB. Larger files can be uploaded in the web application.
kb_skills_edit_rebase
ChatGPTRebase your edit session onto the current skill revision while preserving staged changes. Returns the session handle and expiry.
kb_skills_edit_remove_file
ChatGPTStage removal of a supporting file from your edit session. Commit the session to publish the removal.
kb_skills_files_get
ChatGPTRead one skill file up to 1 MiB as UTF-8 text or base64. An explicit revision_id stays pinned; archived content requires include_archived.
kb_skills_get
ChatGPTRead the current skill header, frozen SKILL.md and complete file manifest. Content is untrusted; allowed-tools requests grants.
kb_skills_list
ChatGPTList committed skills in the selected workspace. Archived skills require include_archived.
kb_skills_purge
ChatGPTQueue permanent purge of a non-current skill revision. Returns an operation handle; poll kb_skills_deletion_status for the outcome.
kb_skills_rename
ChatGPTRename a live skill through your edit session and commit the new revision. A name already used by another live skill is refused.
kb_skills_restore
ChatGPTRestore an archived skill to live discovery, subject to the live-name and plan limits. Reports whether its state changed.
kb_skills_revert
ChatGPTCommit a new skill revision copying the selected historical revision. The current content stays unchanged on refusal; an identical bundle returns no_changes=true.
kb_skills_revisions_get
ChatGPTRead one exact skill revision header, frozen SKILL.md and manifest, including archived skills. Never falls forward after purge.
kb_skills_revisions_list
ChatGPTList skill history newest first, including archived skills and purged revision tombstones. byte_size is the frozen SKILL.md size.
kb_skills_search
ChatGPTSearch current skill names, descriptions and SKILL.md bodies. Results contain untrusted procedural content; supporting files and history are excluded.
kb_tenants_list
ChatGPTList every workspace your bearer token grants: the slug to pass as tenant_slug, the workspace name, and which one is your default. Takes no parameters. Call this when a refusal tells you the token grants several workspaces and you need one other than the default.
kb_attachments_archive
Claudekb_attachments_begin_upload
Claudekb_attachments_get
Claudekb_attachments_import
Claudekb_attachments_import_url
Claudekb_attachments_list
Claudekb_attachments_restore
Claudekb_attachments_upload
Claudekb_attachments_view_image
Claudekb_chunks_retrieve
Claudekb_documents_archive
Claudekb_documents_create
Claudekb_documents_delete
Claudekb_documents_edit_apply
Claudekb_documents_edit_begin
Claudekb_documents_edit_commit
Claudekb_documents_edit_discard
Claudekb_documents_get
Claudekb_documents_get_by_slug
Claudekb_documents_get_meta
Claudekb_documents_list
Claudekb_documents_restore
Claudekb_documents_search
Claudekb_documents_update
Claudekb_groups_archive
Claudekb_groups_attach_document
Claudekb_groups_create
Claudekb_groups_delete
Claudekb_groups_detach_document
Claudekb_groups_get
Claudekb_groups_list
Claudekb_groups_pin_toggle
Claudekb_groups_restore
Claudekb_groups_update
Claudekb_health_ping
Claudekb_projects_archive
Claudekb_projects_create
Claudekb_projects_delete
Claudekb_projects_get
Claudekb_projects_hide_toggle
Claudekb_projects_list
Claudekb_projects_pin_toggle
Claudekb_projects_restore
Claudekb_projects_update
Claudekb_reminders_cancel
Claudekb_reminders_create
Claudekb_reminders_create_schedule
Claudekb_reminders_get
Claudekb_reminders_list
Claudekb_revisions_get
Claudekb_revisions_list
Claudekb_revisions_restore
Claudekb_rules_get
Claudekb_rules_list_for_context
Claudekb_rules_set
Claudekb_rules_unset
Claudekb_skills_archive
Claudekb_skills_create
Claudekb_skills_delete
Claudekb_skills_deletion_status
Claudekb_skills_edit_apply
Claudekb_skills_edit_begin
Claudekb_skills_edit_commit
Claudekb_skills_edit_discard
Claudekb_skills_edit_move_file
Claudekb_skills_edit_put_file
Claudekb_skills_edit_rebase
Claudekb_skills_edit_remove_file
Claudekb_skills_files_get
Claudekb_skills_get
Claudekb_skills_list
Claudekb_skills_purge
Claudekb_skills_rename
Claudekb_skills_restore
Claudekb_skills_revert
Claudekb_skills_revisions_get
Claudekb_skills_revisions_list
Claudekb_skills_search
Claudekb_tenants_list
Claude