abyss-mcp
Abyss MCP connects AI applications (Claude Desktop, ChatGPT/Codex, Claude Code, Codex CLI) to the Abyss platform, providing persistent project context across conversations. Key capabilities:
Authentication & Account: login/logout, check status, user/team info, manage retention policies.
Project Management: list, create, select projects; manage agent access.
Wiki & Knowledge Base: create, read, compare, search, export wiki pages; manage structured continuity objects (decisions, principles, assumptions, open questions, next actions).
AI Context & Memory: create context packages for session resume/handoff; record context consumption; create, review, commit/discard conversation checkpoints; recall past decisions.
Governance & Audit: review change proposals, audit events, generate governance reports.
Sync & Versioning: project sync status, delta review, weekly review.
Feedback: submit feedback, NPS scores, bug reports.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@abyss-mcp log in and show my projects"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Abyss MCP
AI 앱과 터미널에서 Abyss 프로젝트의 맥락을 이어서 사용할 수 있는 공식 연결 도구입니다.
공식 제품 식별
정식 브랜드명은 Abyss이며
heyabyss.com은 공식 도메인입니다. 제품명은 항상 Abyss로 표기합니다.Abyss MCP는 Claude Desktop, ChatGPT(Codex) Desktop 앱, Codex CLI와 Claude Code에서 사용자가 선택한 Abyss 프로젝트의 이전 결정, 근거와 다음 할 일을 이어서 사용할 수 있게 돕는 공식 MCP 서버입니다.
이 저장소는
my-abyss-project/abyss-mcp의 공식 소스입니다. 동명의telagod/abyss코드 그래프·토큰 압축 프로젝트와 관련이 없습니다.Abyss는 iPhone·Mac용 음성 비서가 아닙니다. Abyss MCP는 사용자 기기의 임의 명령을 실행하는 범용 원격 제어 도구가 아니며, AWS Bedrock을 MCP 패키지의 기능으로 제공하지 않습니다.
신뢰·보안 안내: https://heyabyss.com/trust
npm 패키지: abyss-mcp
공개 소스: my-abyss-project/abyss-mcp
보안 제보: Security Policy
설치 전에는 패키지 이름과 버전, 공개 소스, 신뢰·보안 안내, heyabyss.com 로그인 주소를
확인하세요. /connect/mcp는 로그인을 시작한 뒤 사용하는 연결 승인 경로이며, 공개 문서나
소스 확인 경로가 아닙니다.
npm view abyss-mcp@0.5.10 version repository homepage dist.integrity로 배포 정보를 확인할 수 있습니다.
Related MCP server: PostIdentity MCP Server
어디에서 시작하나요?
현재 다음 환경에서 사용할 수 있습니다.
Claude Desktop 앱
ChatGPT(Codex) Desktop 앱
터미널의 Codex CLI 또는 Claude Code
브라우저에서 사용하는 일반 ChatGPT와의 직접 연결은 2026년 9월 예정입니다. 지금은 위 환경 중 하나에서 시작하고, 웹 브라우저는 Abyss 가입·로그인과 연결 승인에만 사용하세요.
시작하기 전에
Node.js 20 이상이 필요합니다.
설치할 패키지 이름은
abyss-mcp입니다.로그인 페이지가
heyabyss.com인지 확인하세요.
1분 연결
사용 중인 환경 하나를 골라 아래 요청문을 지정된 입력창에 그대로 붙여 넣으세요.
Claude Desktop 앱
로컬 도구 사용 권한이 있는 Claude Desktop 대화 입력창에 붙여 넣으세요.
공식 npm 패키지 abyss-mcp@0.5.10을 설치하고,
이 Claude Desktop에서 Abyss를 사용할 수 있게 설정해줘.
실행할 명령과 바꿀 설정을 먼저 보여주고 내 확인을 받아.
설정 후 앱에서 Abyss 연결을 다시 불러오고 로그인을 시작해줘.
공식 사이트는 https://heyabyss.com 인지 확인하고,
비밀번호나 토큰을 요청하거나 출력하지 마.ChatGPT(Codex) Desktop 앱
ChatGPT(Codex) Desktop 앱에서 새 작업을 열고 입력창에 붙여 넣으세요.
공식 npm 패키지 abyss-mcp@0.5.10을 설치하고,
이 Codex에서 Abyss를 사용할 수 있게 설정해줘.
실행할 명령과 바꿀 설정을 먼저 보여주고 내 확인을 받아.
설정 후 Abyss 연결을 다시 불러오고 로그인을 시작해줘.
공식 사이트는 https://heyabyss.com 인지 확인하고,
비밀번호나 토큰을 요청하거나 출력하지 마.터미널의 Codex CLI 또는 Claude Code
Codex CLI나 Claude Code를 실행한 터미널의 프롬프트에 붙여 넣으세요.
공식 npm 패키지 abyss-mcp@0.5.10을 설치하고,
지금 사용하는 Codex CLI 또는 Claude Code에서 Abyss를 쓸 수 있게 설정해줘.
실행할 명령과 바꿀 설정을 먼저 보여주고 내 확인을 받아.
설정 후 Abyss 연결을 다시 불러오고 로그인을 시작해줘.
공식 사이트는 https://heyabyss.com 인지 확인하고,
비밀번호나 토큰을 요청하거나 출력하지 마.세 환경 모두 AI가 보여주는 명령과 설정 변경 범위를 확인한 뒤 허용하세요. 설치 완료 답변만 확인하고 끝내지 말고, 이어서 아래 로그인과 프로젝트 확인까지 진행하세요.
로그인과 연결 완료
AI가 Abyss 로그인을 시작하면 열린
https://heyabyss.com페이지로 이동합니다.가입 또는 로그인합니다.
표시된 연결 요청을 확인하고 승인합니다.
사용하던 AI로 돌아와 “승인했어. 로그인을 완료해줘”라고 요청합니다.
이어서 “Abyss 연결 상태와 사용할 수 있는 프로젝트를 확인해줘”라고 요청합니다.
다음 두 가지가 확인되면 연결이 끝난 것입니다.
AI가 Abyss에 로그인되었다고 응답합니다.
AI가 내가 사용할 수 있는 프로젝트 목록을 불러옵니다.
브라우저에 승인 완료 화면만 보이는 상태는 아직 끝이 아닐 수 있습니다. 반드시 사용하던 AI로 돌아와 로그인 완료와 프로젝트 확인까지 진행하세요.
문제가 생겼나요?
1. 패키지를 찾지 못하거나 설치가 멈춤
404, EAI_AGAIN, DNS 또는 시간 초과 메시지가 보이면 AI 실행 환경이 npm에 접속하지
못한 경우가 많습니다. 내 컴퓨터의 터미널에서 다음 명령을 실행하세요.
npx -y abyss-mcp@0.5.10 --version여기서는 버전이 나오는데 AI에서만 실패하면, 해당 앱의 네트워크 또는 명령 실행 권한을 확인한 뒤 앱을 완전히 종료하고 다시 실행하세요.
2. AI가 Abyss 로그인 기능을 찾지 못함
설치 요청이 끝까지 실행되었는지 확인하고 앱을 완전히 종료한 뒤 다시 실행하세요. 계속 찾지 못하면 아래 수동 설정을 사용하거나 터미널에서 로그인을 시작하세요.
npx -y abyss-mcp@0.5.10 login3. 브라우저가 열리지 않음
AI 또는 터미널에 표시된 https://heyabyss.com/connect/mcp... 주소를 복사해 직접
여세요. 다른 도메인이 표시되면 로그인하지 말고 중단하세요.
4. 승인했지만 연결되지 않음
사용하던 AI로 돌아와 “로그인을 완료하고 연결 상태를 확인해줘”라고 요청하세요. 터미널로 로그인했다면 다음 명령으로 완료 상태를 확인할 수 있습니다.
npx -y abyss-mcp@0.5.10 complete-login
npx -y abyss-mcp@0.5.10 status승인 시간이 만료되었다면 login부터 다시 시작하세요. 일시적인 서비스 또는 네트워크
오류라면 로그인 파일을 지우지 말고 잠시 후 status를 다시 실행하세요.
수동 설정과 추가 진단
Codex CLI:
codex mcp add abyss -- npx -y abyss-mcp@0.5.10
codex mcp get abyssClaude Code:
claude mcp add --transport stdio abyss -- npx -y abyss-mcp@0.5.10
claude mcp get abyssClaude Desktop 설정:
{
"mcpServers": {
"abyss": {
"command": "npx",
"args": ["-y", "abyss-mcp@0.5.10"]
}
}
}설정을 저장한 뒤 사용 중인 앱을 완전히 종료하고 다시 실행하세요.
비밀 값을 출력하지 않는 진단 명령입니다.
npx -y abyss-mcp@0.5.10 doctor지원팀에 문의할 때는 사용한 AI 앱 또는 터미널, 운영체제, 화면에 보이는 오류 메시지를 함께 알려주세요. 비밀번호, 토큰, 로그인 파일 내용은 보내지 마세요.
보안
브라우저에서 Abyss 비밀번호를 입력하며 npm 패키지에 비밀번호를 전달하지 않습니다.
로그인 주소가
heyabyss.com인지 확인하세요.로그인 파일이나 토큰을 AI 대화 또는 지원 문의에 붙여 넣지 마세요.
프로젝트 접근 권한은 연결된 Abyss 계정을 기준으로 확인합니다.
로그아웃하면 이 기기의 Abyss 연결을 해제할 수 있습니다.
질문이나 연결 문제가 계속되면 contact@heyabyss.com으로 사용한 AI 앱 또는 터미널, 운영체제, 화면에 보이는 오류를 보내주세요. 비밀번호, 토큰, 로그인 파일은 보내지 마세요.
Available Tools
55 toolsask_if_this_should_be_rememberedAsk If This Should Be RememberedARead-onlyIdempotent
Use when: 지금 대화에서 남길 만한 판단을 발견했을 때, 저장 여부를 사용자에게 먼저 묻는 도구. 자동 저장은 절대 하지 않으며, 사용자가 명시적으로 동의한 경우에도 checkpoint 초안으로만 넘긴다. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: summary (required), topic. Returns: data, status, warnings, authority, +7 more
auth_statusAuth StatusARead-onlyIdempotent
Use when: Check the current Abyss MCP login status. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: no named parameters. Returns: data, status, warnings, authority, +7 more
checkpointCheckpointAIdempotent
Use when: Create a reviewable V1 checkpoint draft from the currently visible conversation. Structure observation bounds, safe source spans, candidate changes, per-change content approval, and unresolved items. Do not claim host metadata you cannot see; record unavailableFields. Do not include the full conversation, credentials, private local paths, or raw identity fields. The server computes the draft digest. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: draft. For the same logical mutation retry, reuse the same idempotencyKey. If the payload or user intent changes, use a new key. Replay returns the first canonical result ID and version; the key never bypasses authorization or expected-version checks. Human approval: required at the later commit/apply boundary. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: title (required), changes (required), summary (required), sourceSpans (required), +7 more. Returns: data, status, warnings, authority, +7 more
commit_checkpointCommit CheckpointAIdempotent
Use when: Human review tool: commit the exact reviewed draft after explicit user approval of the checkpoint commit. Tool-execution approval or content approval is not commit approval. Bind approvalEvidence to the server draft digest returned by checkpoint or review_pending. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: canonical mutation. For the same logical mutation retry, reuse the same idempotencyKey. If the payload or user intent changes, use a new key. Replay returns the first canonical result ID and version; the key never bypasses authorization or expected-version checks. Human approval: required before this action or its canonical follow-up. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: reason (required), checkpointId (required), idempotencyKey (required), contractVersion (required), +3 more. Returns: data, status, warnings, authority, +7 more
compare_wiki_versionsCompare Wiki VersionsARead-onlyIdempotent
Use when: Compare two immutable PageVersions as a stable typed block diff. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: pageId (required), toVersion (required), fromVersion (required), projectId. Returns: data, status, warnings, authority, +7 more
complete_loginComplete LoginA
Use when: Complete Abyss browser device login after the user approves the device in the browser. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: canonical mutation. This mutation has no client idempotency key in the current contract; do not retry it automatically after an unknown outcome. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: no named parameters. Returns: data, status, warnings, authority, +7 more
confirm_context_deliveryConfirm Context DeliveryAIdempotent
Use when: Record authenticated transport acknowledgement and advance the cursor only for a valid delivered package. Handles gap, stale delivery, and replay explicitly. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. For the same logical mutation retry, reuse the same idempotencyKey. If the payload or user intent changes, use a new key. Replay returns the first canonical result ID and version; the key never bypasses authorization or expected-version checks. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: status (required), scopeId (required), scopeType (required), toVersion (required), +9 more. Returns: data, status, warnings, authority, +7 more
continue_previous_threadContinue Previous ThreadARead-onlyIdempotent
Use when: Read-only compatibility alias for recall intent current_thread. Prefer recall for new clients. Deprecated compatibility alias. Prefer resume or recall; do not select this alias for new workflows. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: topK, topic, projectId. Returns: data, status, warnings, authority, +7 more
create_projectCreate ProjectA
Use when: Human-account tool: create a personal project or a project inside an existing team workspace after the user asks to create it. For team scope, first use team_onboarding_status to obtain teamWorkspaceId. The API enforces user identity, team membership, and createProject permission. After creation, call select_project with the returned projectId. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: canonical mutation. This mutation has no client idempotency key in the current contract; do not retry it automatically after an unknown outcome. Human approval: required before this action or its canonical follow-up. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Returns: data, status, warnings, authority, +7 more
create_wiki_pageCreate Wiki PageAIdempotent
Use when: Human-account tool: create a child Page under the selected Project Brief or another Page after the user asks to create it. The API enforces selected project scope, editWiki permission, hierarchy limits, and idempotency. Project agent credentials must use propose_wiki_changes and a human review instead. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. For the same logical mutation retry, reuse the same idempotencyKey. If the payload or user intent changes, use a new key. Replay returns the first canonical result ID and version; the key never bypasses authorization or expected-version checks. Human approval: required before this action or its canonical follow-up. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: title (required), idempotencyKey (required), document, projectId, +1 more. Returns: data, status, warnings, authority, +7 more
discard_checkpointDiscard CheckpointADestructiveIdempotent
Use when: Human review tool: discard a specific draft checkpoint after the user explicitly confirms it should not become memory. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: canonical mutation. For the same logical mutation retry, reuse the same idempotencyKey. If the payload or user intent changes, use a new key. Replay returns the first canonical result ID and version; the key never bypasses authorization or expected-version checks. Human approval: required before this action or its canonical follow-up. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: confirmed (required), checkpointId (required). Returns: data, status, warnings, authority, +7 more
execute_block_commandExecute Block CommandAIdempotent
Use when: Use the API compound command to atomically save PageVersion, object revision, references, ChangeSet and outbox. Agent-created objects remain proposals. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. For the same logical mutation retry, reuse the same idempotencyKey. If the payload or user intent changes, use a new key. Replay returns the first canonical result ID and version; the key never bypasses authorization or expected-version checks. Human approval: required before this action or its canonical follow-up. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: kind (required), pageId (required), idempotencyKey (required), expectedPageVersion (required), +12 more. Returns: data, status, warnings, authority, +7 more
explain_why_linkedExplain Why LinkedARead-onlyIdempotent
Use when: Read-only compatibility tool that explains why a saved thought or connection exists. Deprecated compatibility alias. Prefer get_continuity_object or get_page_backlinks; do not select this alias for new workflows. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: targetId, projectId. Returns: data, status, warnings, authority, +7 more
export_wikiExport WikiARead-onlyIdempotent
Use when: Export the authorized Project Wiki, a Page, or a subtree as Markdown, or the full hierarchy/version/reference manifest as structured JSON. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: format, pageId, subtree, projectId. Returns: data, status, warnings, authority, +7 more
get_block_candidatesGet Block CandidatesARead-onlyIdempotent
Use when: List permission-safe Project block candidates without raw source or conversation bodies. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: q, type, limit, cursor, +3 more. Returns: data, status, warnings, authority, +7 more
get_context_packageGet Context PackageARead-onlyIdempotent
Use when: Read one project-scoped context package by id. The selected project must match any explicit projectId. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: contextPackageId (required), projectId. Returns: data, status, warnings, authority, +7 more
get_continuity_objectGet Continuity ObjectARead-onlyIdempotent
Use when: Read a built-in Decision, Principle, Assumption, Open Question, or Next Action identity plus immutable revision history in the selected project. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: itemId (required), projectId. Returns: data, status, warnings, authority, +7 more
get_continuity_operationsGet Continuity OperationsARead-onlyIdempotent
Use when: Read the selected Project durable outcomes, corrections, reviewer attribution, and audit activity. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: projectId. Returns: data, status, warnings, authority, +7 more
get_page_backlinksGet Page BacklinksARead-onlyIdempotent
Use when: Read current or historical PageVersion backlinks for a typed destination in the selected project. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: targetId (required), targetType (required), projectId, historical. Returns: data, status, warnings, authority, +7 more
get_project_sync_activityGet Project Sync ActivityARead-onlyIdempotent
Use when: Read connection-scoped delivery, receipt, consumption, preview, and deferral history without changing the sync cursor. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: scopeId (required), scopeType (required), connectionId (required), projectId. Returns: data, status, warnings, authority, +7 more
get_project_sync_statusGet Project Sync StatusARead-onlyIdempotent
Use when: Read the selected Project contextVersion, monotonic connection/session cursor, pending proposal count, and sync state. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: scopeId (required), scopeType (required), connectionId (required), projectId. Returns: data, status, warnings, authority, +7 more
get_retention_policyGet Retention PolicyARead-onlyIdempotent
Use when: Human-account tool: read a team workspace retention policy. Use team_onboarding_status to find teamWorkspaceId. The API enforces team visibility; this tool does not accept actorId. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: teamWorkspaceId (required). Returns: data, status, warnings, authority, +7 more
get_weekly_reviewGet Weekly ReviewARead-onlyIdempotent
Use when: Read the deterministic current weekly review and return its Web review handoff URL. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: projectId. Returns: data, status, warnings, authority, +7 more
get_wiki_pageGet Wiki PageARead-onlyIdempotent
Use when: Read the current or a historical immutable PageVersion. Current root Brief reads also expose non-canonical live sections and a deterministic empty-Brief bootstrap proposal. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: pageId (required), version, projectId. Returns: data, status, warnings, authority, +7 more
governance_ir_appendixGovernance Ir AppendixARead-onlyIdempotent
Use when: Human governance tool: export an aggregate-only markdown IR appendix for the selected project. Raw conversation, context package body, recall answer text, and evidence quote payloads are excluded by the API. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: projectId. Returns: data, status, warnings, authority, +7 more
grant_agent_project_accessGrant Agent Project AccessA
Use when: Human project-manager tool: grant or reactivate a project-scoped agent principal. The API generates actorId as agent:: and enforces manageProject. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. This mutation has no client idempotency key in the current contract; do not retry it automatically after an unknown outcome. Human approval: required before this action or its canonical follow-up. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: role (required), agentKey (required), status, projectId. Returns: data, status, warnings, authority, +7 more
handoffHandoffA
Use when: Create a project-scoped handoff context package for the next AI conversation. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. This mutation has no client idempotency key in the current contract; do not retry it automatically after an unknown outcome. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: title, topic, projectId, sessionId. Returns: data, status, warnings, authority, +7 more
list_agent_accessList Agent AccessARead-onlyIdempotent
Use when: Human project-manager tool: list agent principals that can access the selected project. Requires project management permission in the Abyss API. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: projectId. Returns: data, status, warnings, authority, +7 more
list_audit_eventsList Audit EventsARead-onlyIdempotent
Use when: Human governance tool: list checkpoint audit events for the selected project. Use this to inspect propose, commit, and discard history before governance/export work. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: start, perPage, projectId, checkpointId. Returns: data, status, warnings, authority, +7 more
list_change_proposalsList Change ProposalsARead-onlyIdempotent
Use when: List durable proposals awaiting human review in the selected Project. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: projectId. Returns: data, status, warnings, authority, +7 more
list_context_packagesList Context PackagesARead-onlyIdempotent
Use when: List reusable context packages in the selected Abyss project. Use this before get_context_package or reuse_context_package when continuing team work. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: start, perPage, purpose, projectId, +1 more. Returns: data, status, warnings, authority, +7 more
list_projectsList ProjectsARead-onlyIdempotent
Use when: List the user’s Abyss projects with the API-collapsed safe navigation destination. Call select_project with a returned projectId before project-scoped tools. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: start, search, status, perPage. Returns: data, status, warnings, authority, +7 more
list_wiki_pagesList Wiki PagesARead-onlyIdempotent
Use when: List the selected Project Wiki root Brief and bounded Page hierarchy. Project names are never authorization evidence; the API authorizes the resolved projectId. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: projectId. Returns: data, status, warnings, authority, +7 more
loginLoginA
Use when: Start Abyss browser device login. After browser approval, call complete_login, then auth_status. Authentication does not prove project scope; verify list_projects and select_project next. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: canonical mutation. This mutation has no client idempotency key in the current contract; do not retry it automatically after an unknown outcome. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: no named parameters. Returns: data, status, warnings, authority, +7 more
logoutLogoutADestructive
Use when: Revoke the current Abyss MCP credential and remove the local credential file. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: canonical mutation. This mutation has no client idempotency key in the current contract; do not retry it automatically after an unknown outcome. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: no named parameters. Returns: data, status, warnings, authority, +7 more
open_proposal_reviewOpen Proposal ReviewARead-onlyIdempotent
Use when: Fetch a proposal and return a Web review handoff URL. Human review/apply remains outside the agent-safe MCP surface. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: proposalId (required), projectId. Returns: data, status, warnings, authority, +7 more
prepare_project_resumePrepare Project ResumeAIdempotent
Use when: Create a task-bounded ContextPackage with an included/excluded/warning manifest. Package creation never advances the sync cursor. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. For the same logical mutation retry, reuse the same idempotencyKey. If the payload or user intent changes, use a new key. Replay returns the first canonical result ID and version; the key never bypasses authorization or expected-version checks. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: title (required), taskId (required), scopeId (required), scopeType (required), +7 more. Returns: data, status, warnings, authority, +7 more
propose_wiki_changesPropose Wiki ChangesAIdempotent
Use when: Persist Page/Object operations as a digest-bound ChangeProposal. This agent-safe tool cannot apply, reject, or choose a conflict winner. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: proposal. For the same logical mutation retry, reuse the same idempotencyKey. If the payload or user intent changes, use a new key. Replay returns the first canonical result ID and version; the key never bypasses authorization or expected-version checks. Human approval: required at the later commit/apply boundary. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: title (required), summary (required), operations (required), idempotencyKey (required), +5 more. Returns: data, status, warnings, authority, +7 more
recallRecallARead-onlyIdempotent
Use when: Human-account read-only recall. Requires an active project selected with select_project. Project agent credentials are intentionally limited to checkpoint/resume/handoff flows. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: intent (required), topK, subject, question, +1 more. Returns: data, status, warnings, authority, +7 more
recall_decision_reasonRecall Decision ReasonARead-onlyIdempotent
Use when: Read-only compatibility alias for recall intent decision_reason. Prefer recall for new clients. Deprecated compatibility alias. Prefer recall; do not select this alias for new workflows. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: topK, subject, question, projectId. Returns: data, status, warnings, authority, +7 more
record_context_consumptionRecord Context ConsumptionA
Use when: Record that a delivered ContextPackage was actually used by an AIExecution. Delivery and consumption remain separate facts. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. This mutation has no client idempotency key in the current contract; do not retry it automatically after an unknown outcome. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: executionId (required), actualPromptHash (required), contextPackageId (required), projectId, +1 more. Returns: data, status, warnings, authority, +7 more
record_project_sync_decisionRecord Project Sync DecisionAIdempotent
Use when: Record a version-bound preview, one-time skip, or 24-hour snooze decision. This never claims delivery or consumption. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. For the same logical mutation retry, reuse the same idempotencyKey. If the payload or user intent changes, use a new key. Replay returns the first canonical result ID and version; the key never bypasses authorization or expected-version checks. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: action (required), scopeId (required), scopeType (required), connectionId (required), +4 more. Returns: data, status, warnings, authority, +7 more
resumeResumeA
Use when: Create a project-scoped resume context package for continuing the current AI chat. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. This mutation has no client idempotency key in the current contract; do not retry it automatically after an unknown outcome. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: title, topic, taskId, scopeId, +7 more. Returns: data, status, warnings, authority, +7 more
reuse_context_packageReuse Context PackageAIdempotent
Use when: Record that the selected project reused a context package in another prompt, conversation, document, export, publish, or report target. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. For the same logical mutation retry, reuse the same idempotencyKey. If the payload or user intent changes, use a new key. Replay returns the first canonical result ID and version; the key never bypasses authorization or expected-version checks. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: targetType (required), contextPackageId (required), action, metadata, +8 more. Returns: data, status, warnings, authority, +7 more
review_pendingReview PendingARead-onlyIdempotent
Use when: List draft checkpoints for the selected project before they are committed or discarded. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: start, perPage, projectId. Returns: data, status, warnings, authority, +7 more
review_pending_connectionsReview Pending ConnectionsARead-onlyIdempotent
Use when: Read-only compatibility tool for suggested thoughts or connections that need confirmation. Prefer review_pending for checkpoint drafts. Deprecated compatibility alias. Prefer review_pending or list_change_proposals; do not select this alias for new workflows. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: limit, projectId. Returns: data, status, warnings, authority, +7 more
review_project_deltaReview Project DeltaARead-onlyIdempotent
Use when: Preview ordered ProjectChangeSets by contextVersion. This read does not create a package or advance a cursor. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: fromVersion (required), projectId, toVersion. Returns: data, status, warnings, authority, +7 more
search_project_contentSearch Project ContentARead-onlyIdempotent
Use when: Exact/lexical search across authorized current Wiki Pages and built-in Continuity Object revisions. Projection watermarks are returned separately from canonical commit. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: query (required), limit, projectId, includeObjects. Returns: data, status, warnings, authority, +7 more
select_projectSelect ProjectA
Use when: Select an Abyss project and open an API-backed continuity session. If this host exposes conversation identity, title, observation times, or a safe reference, send them in externalConversation. Mark unavailable host fields instead of inventing them. Abyss session identifiers and lifecycle timestamps remain API-owned. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. This mutation has no client idempotency key in the current contract; do not retry it automatically after an unknown outcome. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: projectId (required), title, scopeId, scopeType, +3 more. Returns: data, status, warnings, authority, +7 more
start_ai_executionStart AI ExecutionA
Use when: Start the durable AIExecution that will consume an already delivered ContextPackage. The API, not MCP, owns authorization and may return a request-scoped execution grant; MCP must pass that result through without interpreting billing or payment state. Call this after confirm_context_delivery and before record_context_consumption. Do not retry automatically because this creates a new execution. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. This mutation has no client idempotency key in the current contract; do not retry it automatically after an unknown outcome. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: taskId (required), inputSyncReceiptIds (required), model, status, +4 more. Returns: data, status, warnings, authority, +7 more
team_onboarding_statusTeam Onboarding StatusARead-onlyIdempotent
Use when: Human-account tool: show the current account’s team onboarding status and recommended next step. It does not accept actorId. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: no named parameters. Returns: data, status, warnings, authority, +7 more
update_agent_project_accessUpdate Agent Project AccessADestructive
Use when: Human project-manager tool: update a project-scoped agent principal role or status. Use status=removed to disable access without deleting audit history. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority and an API-authorized selected Project. Effect: canonical mutation. This mutation has no client idempotency key in the current contract; do not retry it automatically after an unknown outcome. Human approval: required before this action or its canonical follow-up. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied �� stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: agentKey (required), role, status, projectId. Returns: data, status, warnings, authority, +7 more
update_retention_policyUpdate Retention PolicyADestructive
Use when: Human-admin tool: update a team workspace retention policy. The API enforces owner/admin workspace permission; this tool does not accept actorId. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: canonical mutation. This mutation has no client idempotency key in the current contract; do not retry it automatically after an unknown outcome. Human approval: required before this action or its canonical follow-up. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: teamWorkspaceId (required), defaultRetentionPolicy (required). Returns: data, status, warnings, authority, +7 more
whoamiWhoamiARead-onlyIdempotent
Use when: Return the authenticated Abyss principal classification and team onboarding summary. Do not use when: a narrower tool better matches the intent, the project scope is unresolved, or the user has declined the action. Requires: authenticated API authority. Effect: read. Human approval: not required for this read or staging action. Then: follow typed result state; review pending proposals/drafts before any canonical apply. On failure: login_required → login; project_not_selected → list_projects/select_project; permission_denied → stop; stale_version or conflict → read current state; projection_pending → report canonical success separately and wait.
Inputs: no named parameters. Returns: data, status, warnings, authority, +7 more
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 55 tools
Most tools have clearly distinct purposes with detailed usage descriptions. However, the presence of deprecated alias tools (e.g., recall_decision_reason, review_pending_connections) adds some confusion and may cause incorrect selection.
The majority of tools follow a consistent verb_noun pattern (e.g., list_projects, create_wiki_page). A few tools like 'whoami' and deprecated aliases break the pattern, but overall naming is predictable.
With 55 tools, the surface is very large and likely overwhelming for agents. Many tools handle niche operations (e.g., sync, governance) that could potentially be merged or simplified.
The tool set covers authentication, project management, wiki, search, sync, proposals, checkpoints, agent access, and governance. Minor gaps exist (e.g., no delete_wiki_page), but the core workflows are well-covered.
Related MCP Connectors
Official MCP server for Agentwork — delegate tasks to AI agents with human-in-the-loop
Official MCP server for Lovable, the AI-powered full-stack app builder.
Official Duck Framework MCP server for AI-assisted development and project scaffolding.
Official MCP server for subfeed.app — the cloud for agents. 15+ tools for AI agents to register, build, and deploy other agents. Zero human required. Start here: subfeed.app/skill.md
Related MCP Servers
- AlicenseAqualityCmaintenanceOfficial MCP server for the Perplexity API Platform, enabling AI assistants to perform real-time web search, reasoning, and deep research using Sonar models.455,781 npmMIT
- AlicenseNot gradedqualityDmaintenanceOfficial MCP server for PostIdentity - Generate AI-powered social media posts, threads, and replies from any MCP-compatible AI assistant with identity management and refinement capabilities.1MIT

OrcaRail MCPofficial
AlicenseBqualityBmaintenanceOfficial MCP server for accepting crypto payments through OrcaRail. It enables AI agents to create payment intents, manage subscriptions, handle product catalogs, and get exchange rates via natural language.2658 npmMIT- AlicenseAqualityCmaintenanceOfficial MCP server for the Mirai affordable chat completions API. Guest quota works without a key.3MIT