Available Tools

The active Hamster MCP Server contract exposes 39 tools in full user sessions and eight tools in hosted delivery sessions. The full scope contains 35 CLI/MCP pairs and four MCP-only tools.

Two Access Scopes

Full scope uses your signed-in account and permissions. Your MCP client receives 39 account-scoped read and write tools.
Hosted delivery scope is bound to one delivery job and one delivery resource — a Brief or Task. The agent receives eight tools and cannot list or read unrelated workspace data.

Full Scope: 39 Tools

Search

Tool	What it does
`search`	Find Briefs, notes, Initiatives, Goals, documents, and Tasks by name or task display ID.

Accounts and Members

Tool	What it does
`list_accounts`	List accounts available to you.
`switch_account`	Select the active account for this MCP session.
`list_members`	List members of the active account.

Briefs and Workspace Questions

Tool	What it does
`list_briefs`	List Briefs in the active account.
`get_brief`	Get a Brief's content and metadata.
`create_brief`	Create a Brief in the active account.
`update_brief`	Update a Brief's title, description, status, or linked Initiatives.
`ask_hamster`	Start a durable workspace turn, optionally continuing an existing `thread_id`.
`get_hamster_reply`	Read the exact turn identified by `thread_id` and `turn_id`; repeat this read after `retry_after_ms` while a result is pending.

Notes

Tool	What it does
`list_notes`	List document-backed notes.
`get_note`	Get note metadata and optional markdown content.
`create_note`	Create a document-backed note with a title and optional content.

Initiatives

Tool	What it does
`list_initiatives`	List Initiatives with title-and-description search, status and health filters, exclusions, and pagination.
`get_initiative`	Get an Initiative's details and optionally its Markdown content.
`create_initiative`	Create an Initiative with a title and description.
`update_initiative`	Update an Initiative's title, description, or status.
`archive_initiative`	Archive an Initiative so it no longer appears in lists by default; Initiatives cannot be hard-deleted.
`unlink_brief_from_initiative`	Remove a Brief's link to an Initiative.

Tasks and Subtasks

Tool	What it does
`list_tasks`	List Tasks for a Brief with supported filters.
`get_task`	Get a Task, its subtasks, and dependencies.
`create_task`	Create a Task under a Brief.
`update_task`	Update a Task's title, description, priority, assignee, or other fields.
`update_task_status`	Move a Task to `todo`, `in_progress`, or `done`.
`get_next_task`	Get the next unblocked Task by priority and dependency state.
`list_subtasks`	List a Task's subtasks.
`create_subtask`	Add a subtask to a Task.
`update_subtask`	Update a subtask's content or status.
`delete_subtask`	Remove a subtask.

Documents and Plans

Tool	What it does
`list_documents`	List the account's documents, optionally filtered by document type, parent document, search, or owner.
`get_document`	Get a document and its content.
`get_plan`	Get a Brief's generated Plan and Tasks.
`generate_plan`	Generate a Plan and Tasks for a Brief.
`trigger_delivery`	Start delivery for a Brief.

Goals

Tool	What it does
`list_goals`	List organizational Goals with title-and-description search, status filters, exclusions, and pagination.
`get_goal`	Get a Goal's details.

Context Graph

Tool	What it does
`search_knowledge_graph`	Search the Context Graph by topic and return typed evidence with relevance and confidence.
`explore_entity`	Explore a named entity's relationships, edge types, lanes, and confidence.
`get_neighborhood`	Resolve affinity edges between seed context node IDs and the rest of the graph.

Initiative and Goal List Behavior

list_initiatives combines its optional filters. Search matches the title or description without case sensitivity. Status filters support planned, active, completed, and archived; health filters support not_started, on_track, at_risk, and off_track. Archived Initiatives are omitted by default and included when the status filter requests them. Status exclusions can remove archived or other matching Initiatives.

list_goals also combines its optional filters and searches titles and descriptions without case sensitivity. Status filters support draft, refining, shipping, done, blocked, and archived. Archived Goals are omitted by default and included when the status filter requests them. Status exclusions can remove archived or other matching Goals.

Both list tools use limit and offset pagination, return has_more, default to 50 results, and allow up to 200 results per page. Results use stable newest-updated ordering.

Hosted Delivery Scope: Eight Tools

Tool	Delivery use
`get_brief`	Read the Brief permitted by the delivery resource binding.
`get_plan`	Read the Plan permitted by the delivery resource binding.
`get_task`	Read a Task permitted by the delivery resource binding.
`get_next_task`	Get the next unblocked Task permitted by the delivery resource binding.
`list_tasks`	List Tasks permitted by the delivery resource binding.
`list_subtasks`	List subtasks permitted by the delivery resource binding.
`get_document`	Read a document permitted by the delivery resource binding.
`update_task_status`	Update the status of a Task permitted by the delivery resource binding.

A hosted delivery credential is bound to one delivery job and one Brief or Task, so the agent cannot list or read unrelated workspace data.

Discovery and Authentication

Your MCP client calls tools/list and receives the surface allowed by its credential. Normal OAuth sessions use your signed-in account context and receive all 39 tools. Hosted delivery uses a job-and-resource-bound credential, so the same server returns only the eight hosted delivery tools. Each tool arrives with its input schema and description, which are rich enough that the AI can infer when to use a tool without you spelling it out — "what's my next Task?" maps to get_next_task, "what Initiatives do I have open?" maps to list_initiatives, and so on.

CLI Parity Is Partial

35 of the 39 MCP tools have a corresponding Hamster CLI command. Four are MCP-only: get_hamster_reply, because the CLI polls for a pending reply inside its interactive chat command; and search_knowledge_graph, explore_entity, and get_neighborhood, because the CLI has no Context Graph query surface. The CLI also has 14 deliberate CLI-only commands for local setup and sync, authentication, skill management, document authoring, convenience shortcuts, and binary updates, so do not assume that every CLI command is an MCP tool.

Tips

Use search when you know an artifact name instead of paging through list tools.
Preserve thread_id for a genuine Ask follow-up. Poll a pending turn with get_hamster_reply; do not call ask_hamster again to check status.
Use get_next_task when an agent needs the highest-priority Task with no unfinished dependencies.
Tool calls respect the active account and your role. There is no MCP admin override.
Navigable URLs are embedded in tool responses — the AI can read the direct link to any resource and show it to you.