@abe-ing/mcp — ABE Model Context Protocol server

A stdio MCP server that gives Claude Code (and any MCP-capable agent) typed tools to drive ABE project boards and tasks — so you can run engineering issues inside ABE instead of GitHub, straight from your editor:

"list my ABE boards" · "create a board called Sprint 12 with columns todo, doing, done" · "add a P0 bug to Sprint 12 assigned to [email protected]" · "move task 91 to review"

It wraps the ABE REST API and authenticates with a Personal Access Token.

Setup in Claude Code

1. Generate a Personal Access Token in the ABE web app: Settings → Developer → Generate a token (scope full, or tasks + boards). Copy it — it's shown once.
2. Add the server.

CLI:

claude mcp add --transport stdio \
  --env ABE_TOKEN=abe_pat_xxxx \
  --env ABE_API_URL=https://api.abe.ing \
  abe -- npx -y @abe-ing/mcp

Or .mcp.json (commit to a project; keep the secret out of git via shell expansion):

{
  "mcpServers": {
    "abe": {
      "command": "npx",
      "args": ["-y", "@abe-ing/mcp"],
      "env": {
        "ABE_API_URL": "https://api.abe.ing",
        "ABE_TOKEN": "${ABE_TOKEN}"
      }
    }
  }
}

Environment

| Var | Required | Default | Notes | |-----|----------|---------|-------| | ABE_TOKEN | yes | — | Personal Access Token (abe_pat_…). The server exits if unset. | | ABE_API_URL | no | http://localhost:3000 | ABE API base URL. Use https://api.abe.ing for production. |

There is no tenant setting — a PAT is bound to its workspace at generation time, so the target workspace is whichever you minted the token for.

Tools

| Tool | Purpose | |------|---------| | list_boards | List project boards (id, slug, name, columns). Call first before creating tasks. | | create_board | Create a project board: name, optional columns[], description. | | update_board_columns | Replace a project board's columns: board (slug/name/id), columns[]. | | list_tasks | List tasks. Filters: board, status, type, label, all, include_done, limit. Default = your open tasks. | | get_task | Fetch one task by id, including its body: body_markdown (the linked worksheet doc rendered as markdown — source of truth) and description (create-time text; can be stale). | | get_worksheet | Fetch a worksheet (doc) by id with bodyMarkdown rendered server-side. Use with a task's worksheet_id, or for any other doc. Returns id, title, body, bodyMarkdown, visibility, lastEditedAt, createdByName (never the share token or owner email). | | update_worksheet | Replace a doc's body from markdown (body_markdown, converted server-side, previous body versioned) and/or set title. Whole-doc replace — get_worksheet first if you mean to edit. Last write wins against an open editor. | | create_task | Create a task: board, title, + description, type, priority, size, labels[], assignee_email, due_date, start_date, end_date, estimate, column. | | update_task | Update any property by id (incl. status, move via board). Nullable fields accept null to clear. body_markdown (or description — same meaning as on create) replaces the task doc; the previous body is kept as a version. | | move_task | Move a task to a column. | | complete_task | Mark a task complete. | | assign_task | Assign by assignee_email (resolved server-side). | | add_task_comment | Add a PR-style comment/status note to a task (id, body). Recorded as authored by the agent (origin: "agent"). | | list_task_comments | List a task's comments (id), chronological. | | update_task_comment | Edit a comment (id, comment_id, body) — own comments only (server-enforced). | | delete_task_comment | Delete a comment (id, comment_id) — own comments only. | | get_task_history | Fetch a task's full timeline (id): field/status events + comments + worksheet edits, merged chronologically. | | delete_task | PERMANENTLY delete a task (id). Irreversible — only when the user explicitly asks; otherwise prefer complete_task. | | list_members | List workspace members (admin/full token). Optional — assigning by email works without it. |

Enumerations are enforced by the tool schemas: priority (P0/P1/P2/urgent/high/normal), task_type (task/feature/bug/suggestion/improvement/ops), size (XS/S/M/L/XL), status (open/completed), dates as YYYY-MM-DD. column is free-form (per-board) and validated by the server.

Assignees are resolved by email or name on the server — the model never needs internal user ids. Boards are referenced by slug, name, or id (resolved automatically).

Error handling

Tools never throw at the protocol level. On failure they return a model-recoverable isError result with guidance — e.g. an invalid column tells the model to call list_boards for the valid set; a 401 tells it the token is invalid/revoked; a 402 explains the workspace billing is locked (blocking reads and writes).

Develop

npm install
npm run build      # tsc → dist/
npm test           # vitest
npm run typecheck

Pinned to @modelcontextprotocol/[email protected] (exact). A future SDK major renames the package and moves to zod v4 — do not cross that boundary without rewriting the imports and re-checking registerTool.