Beck MCP server

Read from the package source without running it. The installed server may list more.

• accept_task

  Accept a ready_for_review task (marks completed). Follow autonomy mode: human_review → only if they asked; peer_review → reviewer/PM bot; full → allowed when the job brief said so. Unspecified mode → prefer human review.

• add_comment

  Add a comment to a task. Use this to document progress, ask questions, or provide updates.

• answer_decision

  Answer an open decision: give an answer or keep the default. For peer-review bots, only when the autonomy mode allows it; otherwise the human answers in Beck.

• ask_decision

  Record a choice the human may want a say in, with the default you are going with, then keep working on that default. Returns right away; the answer comes back on a later start_task, finish_task or recommend_task. Set blocking only for hard-to-undo choices (it blocks the affected tasks until answered, so move to other work). FYI notes for the reviewer belong in progress questions instead.

• bind_workspace

  Bind a Beck workspace to this git repo (the directory the agent runs in) so later calls can omit workspaceId. Do this once per repo after the human picks the workspace.

• create_api_tokenwrite action

  Mint a new API token named for a bot (e.g. Coder, Reviewer). Requires a mint-capable coordinator token (Settings → Tokens). Returns the raw secret once — hand it only to that bot's private config. Worker tokens cannot mint.

• create_documentwrite action

  Create a markdown document in a workspace.

• create_epicwrite action

  Create an epic in a workspace. Use create_tasks to add its work.

• create_taskswrite action

  Add tasks to an epic, in order, at the bottom of the backlog (or icebox). Use when scope grows mid-run or to finish a partial record_plan.

• create_workspacewrite action

  Create a workspace in the current org or personal account. Default: only when the human asked for a new workspace. If they named an existing one, bind to that instead. An empty bound workspace gets a prd document, not a second workspace. The human's prompt overrides this default.

• delete_documentwrite action

  Delete a document. etag is required (If-Match from a prior read_document) so concurrent deletes cannot race a newer edit.

• finish_taskwrite action

  Submit a task for review. Changes status to 'ready_for_review'. Pass progress to log what you did and finish in one call. Next step is accept_task / reject_task under the active autonomy mode (human_review, peer_review, or full).

• get_epic

  Get an epic: working spec, last-plan provenance, imported tasks, GitHub link, and latest suggested breakdown. Use after plan_it or when you need the spec without a task id.

• get_epic_github

  Get the GitHub branch and PR URL linked to an epic.

• get_task

  Get full details of a task including description, acceptance criteria, and comments.

• get_task_context

  Get rich context for a task including epic spec, related tasks, comments, and activity. Use this when starting work on a task to understand the full context.

• import_suggested_tasks

  Import pending Plan-it / breakdown suggestions from an epic into the backlog or icebox. Use when asked or when autonomy mode allows. Prefer get_epic first to see pending indices.

• link_epic_github

  Link a GitHub branch or PR to an epic. Use this after creating a branch or opening a PR for an epic's work.

• list_api_tokens

  List API tokens for this account in the current org (names, prefixes, kind, mint permission). Never returns secrets. Use when staffing multi-bot channels.

• list_comments

  List all comments on a task.

• list_decisions

  List decisions in the workspace: open ones waiting on the human (costly first), then answered ones, newest first.

• list_document_versions

  List saved versions of a document (id, time, actor, etag). Use restore_document to roll back.

• list_documents

  List planning documents in a workspace. Documents are the living PRD/notes corpus.

• list_epics

  List epics in a workspace (title, status, task counts, last plan, GitHub).

• list_progress

  List structured progress updates for a task.

• list_tasks

  List tasks in a workspace. Can filter by status, epic, or show icebox items.

• list_workspaces

  List Beck workspaces this token can see. Tokens are org-scoped — this is not a default workspace. If the human named one, use only that id. If more than one and none named, ask. create_workspace only if they asked for a new one.

• log_progress

  Log a structured progress update for a task. Use this to document what you attempted, what succeeded, what failed, and any blockers or questions.

• plan_it

  For people planning without an agent: snapshot selected documents into one or more epics, optionally with Beck's AI task breakdown. If you planned the work yourself, use record_plan instead. Suggested tasks are not in the backlog until import_suggested_tasks (or a human imports in Beck).

• read_document

  Read a workspace document by path (e.g. prd or research/notes).

• recommend_task

  Get recommendations for what task to work on next. Returns tasks prioritized by: 1) Your in-progress work, 2) Tasks assigned to you, 3) High-priority unassigned tasks.

• record_plan

  Record an approved plan in one call: saves the plan markdown as a document under plans/, then creates each epic and its tasks in order. Use right after the human approves a plan (for example when leaving plan mode) if it has 3 or more tasks. You do the breakdown; Beck does not re-plan it.

• reject_task

  Reject a ready_for_review task back to started. reason is required. Follow the same autonomy mode rules as accept_task.

• restore_document

  Restore a document to a previous version. etag is the current document etag (If-Match), not the version's etag.

• resume

  Brief for picking work back up in this repo's workspace: tasks in progress, waiting for review, next up, open epics and plan documents. Call at session start and after the conversation is compacted, before re-reading code.

• revoke_api_token

  Revoke an API token by id. Owner only via API path.

• set_relevant_files

  Set the relevant files for a task. Use this to record which files are important for the task, so future work sessions know where to look. Paths must be relative to project root.

• start_task

  Start working on a task. This assigns the task to you and changes status to 'started'. Use this before beginning implementation work.

• update_taskwrite action

  Update task details (title, type, points, tags, epic, description, blocked status, acceptance criteria, relevant files). Status cannot be set here — use start_task, finish_task, accept_task, or reject_task.

• whoami

  Return the identity behind this API token (token name/kind, mint permission, owner). Use to know how comments and progress will be attributed.

• write_documentwrite action

  Update a document. etag is required (If-Match from a prior read_document). Writes without an etag are rejected so concurrent edits cannot silently overwrite.

• Code scan23 source files scanned25/25
• –Live reliabilityno gateway calls yet and no remote to proben/a
• –Tool poisoningtools not inspected (local package is not executed); not countedn/a
• Auth qualitystatic API keys via environment variables6/15
• Maintenancelast push 1 days ago15/15
• Maintainer identitynamespace and repository owner differ; website matches verified namespace6/10

Safety score

80/100

higher than 60% of 48,255 servers

Tools

41

the typical server has 7

Tools that change things

22%

typical: 0%. Fewer write tools means less an agent can break

Downloads a week

453

more than 87% of 12,191 servers with a package

Beck: common questions

Is Beck MCP server safe?

Mostly: it is graded B (80/100). Read the Beck safety report

How do I install Beck?

It runs on your machine. Copy the Claude Code, Cursor, VS Code or Claude Desktop config from the install section.

Does Beck need an API key?

Yes. The registry entry asks for BECK_API_TOKEN.

Is Beck maintained?

The last commit was 37 days ago (2026-09-28). The latest release is v1.6.1.

How does Beck compare with other MCP servers?

safety score 80/100 (higher than 60% of 48,255 servers); tools 41 (the typical server has 7); tools that change things 22% (typical: 0%. Fewer write tools means less an agent can break). Full comparison.