GC AI

List or fetch playbooks. Pass `id` to fetch one playbook including its full `guide` text. Otherwise returns a page of playbooks you can access (those you created, org-visible, shared with you, and official ones); narrow with `q` and page with `cursor`/`limit`. Always returns a `playbooks` array (empty when nothing matches) and a `next_cursor` (null on the last page).

get_playbooks

List or fetch projects. Pass `id` to fetch one project. Otherwise returns a page of projects you can access (those you created, open projects, and ones shared with you); narrow with `q` and page with `cursor`/`limit`. Always returns a `projects` array (empty when nothing matches) and a `next_cursor` (null on the last page).

get_projects

List or fetch skills. Pass `id` to fetch one skill including its full `content`. Otherwise returns a page of skills you can access (those you created, org-visible, shared with you, and official ones) as summaries without `content`; narrow with `q` and page with `cursor`/`limit`. Always returns a `skills` array (empty when nothing matches) and a `next_cursor` (null on the last page).

get_skills

List or fetch Contract Intelligence vaults. A vault holds a set of contracts that GC AI has read, with the extracted fields the team tracks. Pass `id` to fetch one vault. Otherwise returns a page of the vaults you can access; narrow with `q` and page with `cursor`/`limit`. Always returns a `vaults` array (empty when nothing matches) and a `next_cursor` (null on the last page). Use a returned `id` as `vault_id` when adding a document with `upload_file` or `start_file_upload`, or to ground `ask_gcai` in that vault's documents. Archived vaults are not listed. `document_count` is the number of documents that have finished scanning and can lag a fresh upload; confirm an upload with `get_files({ id })`, not with this count.

get_vaults

Run a playbook (by `id`) against uploaded files (`file_ids`) and get structured check results. Files must be uploaded and `ready` first; the playbook must be `completed`. Pass `check_ids` to run a subset of checks, or omit to run all. Returns a job envelope, not final results: while `status` is `pending` or `running`, keep polling `run_playbook_status` with the returned `job_id` until `status` is `succeeded` (then `result` holds the check results).

run_playbook

Save an `ask_gcai` chat into GC AI history and return a shareable deep link. API chats stay out of history until saved. Pass `project_id` to also file the chat into a project you can edit. Idempotent: saving a chat that is already saved (and already in that project) returns the same link with no side effects.

ask_gcai_save_chat

Start uploading a file that only exists in your local/sandbox filesystem (use this when no host-attached file reference is available, e.g. on Claude). Returns a short-lived signed `upload_url`: PUT the file bytes straight to it from your sandbox, for example `curl -X PUT --upload-file <path> "<upload_url>"`. Then call `finish_file_upload` with the returned `file_id` to register and process the file. Target with `project_id`, `vault_id`, or `scope` (mutually exclusive, default "my-files"). If the host already populates a file reference, prefer `upload_file` instead.

start_file_upload

Update a playbook you can write to. Only the fields you pass change. Pass `checks` to replace the whole check set (read it first, edit, send it back); omit `checks` to leave them. Returns the playbook and its checks.

update_playbook

Update a project you can write to. Only the fields you pass change; omit a field to leave it, or pass null to `description`/`custom_instructions` to clear it. Membership is managed with the attach/detach tools, not here.

update_project

Update a skill you can write to. Only the fields you pass change; omit a field to leave it. `file_ids` replaces the whole attachment set. Visibility is not editable here — manage skill sharing in the GC AI app.

update_skill

Upload a file the user has attached into GC AI. To upload, just call this tool: the host populates the `file` argument automatically. Do not read or pass a local/sandbox path (e.g. /mnt/data/...) and do not construct `file.download_url` yourself. The bytes are fetched, stored, and processed (text extraction + embeddings) in place. Poll readiness with `get_files({ id })` until `status` is "ready" (then use the file) or "failed" (then `extraction_error` explains why). Target the upload with `project_id` (into a project), `vault_id` (as a document in a Contract Intelligence vault), or `scope` ("my-files" / "organization"); the three are mutually exclusive and default to "my-files".

upload_file