2ools MCP Server
OfficialThis server has a remotely hosted endpoint. Open its connector listing for connection details, authentication requirements, and setup instructions:
2ools MCP Server
Remote MCP server for 2ools — build, version, review and export websites, web apps and games from inside the AI chat you already use.
https://2ools.app/mcpYour assistant writes the site. 2ools is where it lives: every result becomes a restorable version, previews are shareable links that need no account to open, and the source exports as plain files with no proprietary runtime.
One tool is free and needs no account at all — create_public_preview turns a
standalone HTML document your model just wrote into a real, viewable URL. The rest
connect over OAuth 2.1 with PKCE and least-privilege scopes.
Quick start
Paste this into the AI client you want to connect:
Connect this client to 2ools (github.com/2ools/mcp-server, MCP endpoint https://2ools.app/mcp). Retrieve https://2ools.app/start non-visibly with an HTTP or web-fetch tool; never navigate the in-app browser to that instructions page. Use this client's in-app browser for every 2ools sign-in and authorization page; never use Chrome, another external browser, or the system browser when this client has an in-app browser or preview pane. The exact OAuth approval URL must be the first 2ools page you visibly open. Before I must act, your only user-facing setup message may be
Connecting 2ools in this client's browser…. Do not output insight blocks or narrate setup mechanics. Ask me exactly once to sign in and click Authorize. Do not end your turn while authorization is pending; after it succeeds, keep the 2ools app visibly open in that pane. Fetch https://2ools.app/start and follow its "For your AI" steps exactly.
The live start page is the single current connection ritual for every client. It detects the current client path, keeps OAuth alive, and leaves only sign-in and the Authorize click to the person. Manual MCP configuration is a fallback only for clients that cannot manage their own remote-MCP setup.
After authorization, the initiating context selects the app destination: an exact project's Project OS when its ID is known, the new-project or existing-project workspace flow when that intent is known, and the authenticated MCP workspace for a generic install. The public homepage is never the fallback.
Registry introspection proxy
Registries that require a local stdio process can run the included credential-free proxy. It forwards MCP JSON-RPC traffic to the same verified remote endpoint; it does not mock or reimplement any 2ools tools. It translates the stdio handshake into the remote endpoint's current per-request MCP envelope and returns the live schemas from 2ools.
npm ci
npm startThe included Dockerfile runs the same dependency-free proxy for reproducible
registry introspection. It intentionally does not perform an interactive OAuth
flow. Clients using authenticated workspace tools should connect directly to
https://2ools.app/mcp so OAuth remains end-to-end with 2ools.
2ools Skill
For assistants that support repository skills, add or copy the 2ools Skill. It keeps the temporary-preview and OAuth workspace boundaries explicit. Client support for remote MCP and Skill folders varies; the live start page routes each client to its supported flow.
Try it with no account
Ask your assistant:
Write this as one standalone HTML document, create a free 2ools preview, and show me the link.
You get back a real URL that anyone can open. It expires after 24 hours.
Related MCP server: OffByOne Runtime MCP Server
What you can do with it
Ask your assistant… | It uses |
"Open this generated site in 2ools and give me a client-safe preview link" |
|
"Save this conversation as a durable 2ools project" |
|
"Make the hours bigger and show me the new version" |
|
"Compare the last two versions" |
|
"What changed on this project since Tuesday?" |
|
"Export the source as a zip" |
|
Tools
43 tools, each with explicit readOnlyHint and destructiveHint annotations so a
client can reason about blast radius before calling anything.
Free — no account required
Tool | Effect | What it does |
| write | Turn a complete standalone HTML document already created in this chat into a temporary, read-only 2ools preview. |
Build and refine
Tool | Effect | What it does |
| write | Create a new 2ools project from a structured conversation brief and queue one durable build. |
| write | Create one protected child version from a saved base version and concrete review feedback. |
| read-only | Get one build's current status, progress, errors, completed version IDs, and protected preview URLs. |
| destructive | Cancel a queued or running non-Agent 2ools build. |
| read-only | Compare two saved versions in one project without returning source. |
| read-only | List the Curators, Rulebooks, and Skills this paid account may explicitly apply. |
Read a project
Tool | Effect | What it does |
| read-only | List recent 2ools projects the authenticated account can access, including owned and shared projects. |
| read-only | Get a 2ools project's brief, selected pages, active version, and compact version history. |
| read-only | Read one source-free operating packet before acting: the active version, exact frozen Curator snapshot, revision-… |
| read-only | Render an interactive, read-only card for one saved 2ools version. |
| read-only | Read the attributable project timeline across human work, external MCP agents, builds, versions, feedback, and… |
| read-only | Wait up to 20 seconds for attributable project events newer than a head_cursor or prior next_cursor. |
Review and approval
Tool | Effect | What it does |
| read-only | List exact saved versions waiting for a separately delegated reviewer. |
| destructive | Use separately delegated review authority to approve one saved version and make it the project's active version. |
| destructive | Use separately delegated review authority to reject one saved version. |
Feedback and shared notes
Tool | Effect | What it does |
| read-only | Read page-anchored review feedback for versions in a project, with explicit human or external-agent attribution. |
| write | Add an attributable review comment to a saved project version. |
| destructive | Owner-only review decision that resolves or reopens one saved feedback item. |
| read-only | Read shared workspace notes and http(s) links, with explicit attribution. |
| write | Add an attributable note or safe http(s) link to the shared project workspace. |
| destructive | Remove one shared workspace pin. |
Export and sharing
Tool | Effect | What it does |
| read-only | Create a five-minute download URL for an exact saved version. |
| read-only | Read the durable sharing state and current URL for one exact saved version. |
| write | Create or recover a durable public or code-protected link for one exact saved version. |
| destructive | Disable every existing public or code-protected URL for one exact saved version. |
Artifacts
Tool | Effect | What it does |
| read-only | List durable project artifacts and each latest immutable version without returning stored bytes. |
| read-only | Read one artifact's immutable version history, provenance, attribution, digests, and review states without… |
| write | Create a five-minute PUT URL for exact bytes. |
| read-only | Create a five-minute download URL for one immutable artifact version. |
| read-only | List exact artifact versions awaiting or carrying a human review decision. |
| write | Approve one pending immutable artifact version and make it current. |
| write | Reject one pending immutable artifact version with useful feedback. |
Agents
Tool | Effect | What it does |
| read-only | List the authorizing account's built-in Agents assigned to one readable project. |
| read-only | Read one Agent's saved Limits, trigger, immutable revision summaries, and redacted append-only run traces. |
| write | Queue real execution for a project-linked Agent the person already test-ran and armed. |
| destructive | Stop one queued or running Agent execution and close its durable run history. |
Work assignments
Tool | Effect | What it does |
| read-only | List durable project work contracts, bounded attempts, leases, budgets, and result references. |
| write | Create a durable assignment with acceptance criteria and hard step, time, and spend ceilings. |
| write | Atomically claim one open assignment and receive a five-minute renewable lease plus an immutable budget snapshot. |
| write | Renew the authenticated worker's active lease without widening its immutable step, time, or spend ceilings. |
| write | End the authenticated worker's lease and submit typed result references for a separate review decision. |
| write | Complete submitted work or reopen it with useful feedback. |
Authentication
Access | Method | Scopes |
| none — genuinely authless | — |
Everything else | OAuth 2.1 authorization code + PKCE (S256) | least-privilege, consented per connection |
The authorization server publishes RFC 8414 and RFC 9728 discovery documents at
/.well-known/oauth-authorization-server and /.well-known/oauth-protected-resource,
supports RFC 7591 dynamic client registration, rotates refresh tokens with family
revocation on replay, and stores only token digests.
Review authority is deliberately separate from build authority. projects:build
lets an agent produce candidate versions; approve_version and reject_version
require projects:approve, which is excluded from default grants. An agent that can
build cannot sign off on its own work unless a human explicitly delegates that.
Design notes
Source never travels through the model transcript. Exports and artifact downloads are five-minute signed URLs bound to the action, the account and the exact version. The MCP response carries metadata and links, not bytes — so a large site does not consume the context window, and generated source is not replayed into a provider log.
Shared previews run in an opaque-origin sandbox. A shared page is served under a
CSP sandbox with connect-src 'none', so a site built by a model cannot read the
viewer's cookies, storage, or reach the network.
External agent output enters review, not production. An artifact uploaded by an outside agent is stored immutably and marked pending; it cannot become the current version until a human with review authority approves that exact digest.
Interactive previews use MCP Apps. render_project_preview returns a real
interactive card via ui://2ools/project-preview-v1.html rather than a wall of text.
Links
Product — https://2ools.app
Connection guide — https://2ools.app/connect-ai
Free website audit (28 deterministic checks, no account) — https://2ools.app/website-audit
Benchmarks: 36 real sites scored — https://2ools.app/website-audit/benchmarks
What 2ools does — https://2ools.app/product
Pricing — https://2ools.app/pricing
LLM guide — https://2ools.app/llms.txt
Privacy & terms — https://2ools.app/legal
Registry
Published to the official Model Context Protocol registry under the
domain-verified namespace app.2ools/mcp.
curl "https://registry.modelcontextprotocol.io/v0/servers?search=2ools"Status
The endpoint is live in production. Health, including the exact deployed release:
curl https://2ools.app/api/healthLicense
MIT — see LICENSE. This repository holds the server's public manifest and documentation; the 2ools application itself is a hosted product.
Available Tools
43 toolsadd_project_feedbackAdd project feedbackBInspect
Add an attributable review comment to a saved project version. This does not modify source or approve the version.
Inputs: comment (required), project_id (required), version_id (required), idempotency_key (required), +5 more.
add_project_pinAdd workspace pinCInspect
Add an attributable note or safe http(s) link to the shared project workspace.
Inputs: kind (required), project_id (required), idempotency_key (required), x, +5 more.
approve_artifact_versionApprove exact artifact versionAIdempotentInspect
Approve one pending immutable artifact version and make it current. Repeating the same decision is idempotent.
Inputs: request_id (required), note.
approve_versionApprove a 2ools versionADestructiveIdempotentInspect
Use separately delegated review authority to approve one saved version and make it the project's active version. Does not publish, export, or change project permissions.
Inputs: project_id (required), version_id (required), confirmation (required), decision_note.
build_from_conversationBuild conversation in 2oolsAIdempotentInspect
Create a new 2ools project from a structured conversation brief and queue one durable build. Returns project_id and job_id; use get_build to retrieve completion status and the protected preview.
Inputs: project_name (required), idempotency_key (required), conversation_summary (required), model, +12 more.
cancel_agent_runCancel governed Agent runADestructiveIdempotentInspect
Stop one queued or running Agent execution and close its durable run history. Completed, failed, stalled, and already-canceled runs keep their terminal truth.
Inputs: job_id (required), run_id (required), agent_id (required), project_id (required).
cancel_buildCancel 2ools buildADestructiveIdempotentInspect
Cancel a queued or running non-Agent 2ools build. Completed versions are not deleted and terminal builds are left unchanged. Governed Agent execution uses cancel_agent_run.
Inputs: job_id (required).
claim_work_assignmentClaim bounded project workAIdempotentInspect
Atomically claim one open assignment and receive a five-minute renewable lease plus an immutable budget snapshot.
Inputs: assignment_id (required), idempotency_key (required), input_refs.
compare_versionsCompare 2ools versionsARead-onlyIdempotentInspect
Compare two saved versions in one project without returning source. Reports lineage, review state, page additions/removals, byte sizes, and SHA-256 identity for each page.
Inputs: project_id (required), left_version_id (required), right_version_id (required).
create_public_previewCreate free 2ools previewAIdempotentInspect
Turn a complete standalone HTML document already created in this chat into a temporary, read-only 2ools preview. Free and authless; expires after 24 hours. This tool does not generate the page—write the HTML first, then call it.
Inputs: html (required), name (required), idempotency_key (required).
create_work_assignmentCreate bounded project workAIdempotentInspect
Create a durable assignment with acceptance criteria and hard step, time, and spend ceilings.
Inputs: brief (required), title (required), project_id (required), idempotency_key (required), +6 more.
decide_work_assignmentReview submitted project workBIdempotentInspect
Complete submitted work or reopen it with useful feedback. Requires separately delegated work-review authority.
Inputs: decision (required), assignment_id (required), idempotency_key (required), note.
get_agentInspect a governed 2ools AgentARead-onlyIdempotentInspect
Read one Agent's saved Limits, trigger, immutable revision summaries, and redacted append-only run traces. Raw prompts, tool arguments, credentials, and source bytes are not returned.
Inputs: agent_id (required), run_limit.
get_artifactGet artifact historyARead-onlyIdempotentInspect
Read one artifact's immutable version history, provenance, attribution, digests, and review states without returning stored bytes.
Inputs: project_id (required), artifact_id (required).
get_buildGet 2ools buildARead-onlyIdempotentInspect
Get one build's current status, progress, errors, completed version IDs, and protected preview URLs. Does not return generated source code.
Inputs: job_id (required).
get_projectGet 2ools projectARead-onlyIdempotentInspect
Get a 2ools project's brief, selected pages, active version, and compact version history. Does not return generated source code.
Inputs: project_id (required).
get_project_activityGet 2ools project activityARead-onlyIdempotentInspect
Read the attributable project timeline across human work, external MCP agents, builds, versions, feedback, and workspace pins. Actor kind is explicit; an external agent is never presented as the human account that authorized it.
Inputs: project_id (required), limit, cursor.
get_project_contextGet 2ools project operating contextARead-onlyIdempotentInspect
Read one source-free operating packet before acting: the active version, exact frozen Curator snapshot, revision-safe Rulebook context, review attention, artifact index, work queue, and current activity cursor. Historical Rulebook text is never guessed after its revision changes.
Inputs: project_id (required), include_governance_text.
heartbeat_work_sessionRenew a work-session leaseAIdempotentInspect
Renew the authenticated worker's active lease without widening its immutable step, time, or spend ceilings.
Inputs: session_id (required), used_steps, used_spend_cents.
list_approval_requestsList 2ools approval requestsARead-onlyIdempotentInspect
List exact saved versions waiting for a separately delegated reviewer. Returns project/version identity and attribution, never generated source.
Inputs: limit, status.
list_artifact_approval_requestsList artifact review requestsARead-onlyIdempotentInspect
List exact artifact versions awaiting or carrying a human review decision. Requires separately delegated artifact review authority.
Inputs: status, project_id.
list_artifactsList project artifactsARead-onlyIdempotentInspect
List durable project artifacts and each latest immutable version without returning stored bytes.
Inputs: project_id (required).
list_build_systemsList 2ools build systemsARead-onlyIdempotentInspect
List the Curators, Rulebooks, and Skills this paid account may explicitly apply. Use before build_from_conversation when the user names a creative system or asks what is available.
Inputs: output_target.
list_project_agentsList project AgentsARead-onlyIdempotentInspect
List the authorizing account's built-in Agents assigned to one readable project. This reveals status and summary, not full Limits or run traces.
Inputs: project_id (required).
list_project_feedbackList project feedbackARead-onlyIdempotentInspect
Read page-anchored review feedback for versions in a project, with explicit human or external-agent attribution.
Inputs: project_id (required), include_resolved.
list_project_pinsList workspace pinsBRead-onlyIdempotentInspect
Read shared workspace notes and http(s) links, with explicit attribution.
Inputs: project_id (required).
list_projectsList 2ools projectsARead-onlyIdempotentInspect
List recent 2ools projects the authenticated account can access, including owned and shared projects. Returns compact metadata without generated source.
Inputs: limit.
list_work_assignmentsList project work assignmentsCRead-onlyIdempotentInspect
List durable project work contracts, bounded attempts, leases, budgets, and result references.
Inputs: project_id (required), status.
prepare_artifact_downloadPrepare exact artifact downloadARead-onlyInspect
Create a five-minute download URL for one immutable artifact version. Stored bytes never enter MCP model output.
Inputs: project_id (required), artifact_id (required), artifact_version_id (required).
prepare_artifact_uploadPrepare reviewed artifact uploadAInspect
Create a five-minute PUT URL for exact bytes. External-agent output always enters pending review and never replaces the current artifact version before approval.
Inputs: kind (required), name (required), sha256 (required), filename (required), +6 more.
prepare_version_exportPrepare portable version exportARead-onlyInspect
Create a five-minute download URL for an exact saved version. Source is delivered by the download route, never embedded in MCP model output.
Inputs: project_id (required), version_id (required).
refine_projectRefine a 2ools projectAIdempotentInspect
Create one protected child version from a saved base version and concrete review feedback. Preserves the base source as currentCode, consumes one persisted project iteration, and returns an asynchronous job_id.
Inputs: feedback (required), project_id (required), base_version_id (required), idempotency_key (required), +1 more.
reject_artifact_versionRequest artifact changesAIdempotentInspect
Reject one pending immutable artifact version with useful feedback. Repeating the same decision is idempotent.
Inputs: note (required), request_id (required).
reject_versionReject a 2ools versionADestructiveIdempotentInspect
Use separately delegated review authority to reject one saved version. If it is active, restores its saved parent as active; the rejected version remains in immutable history.
Inputs: project_id (required), version_id (required), confirmation (required), decision_note (required).
remove_project_pinRemove workspace pinCDestructiveIdempotentInspect
Remove one shared workspace pin. The operation is idempotent.
Inputs: pin_id (required), project_id (required).
render_project_previewShow 2ools project previewARead-onlyIdempotentInspect
Render an interactive, read-only card for one saved 2ools version. Use after get_build or get_project when the user wants to see, open, or share the result. Returns protected preview and access-checked workspace links without source code.
Inputs: project_id (required), version_id (required).
resolve_project_feedbackResolve project feedbackBDestructiveIdempotentInspect
Owner-only review decision that resolves or reopens one saved feedback item.
Inputs: project_id (required), feedback_id (required), resolved.
run_project_agentRun an armed project AgentAIdempotentInspect
Queue real execution for a project-linked Agent the person already test-ran and armed. The immutable saved Limits, project access, account fuse, and online Engine requirement are enforced server-side before a job exists.
Inputs: task (required), model (required), agent_id (required), project_id (required), +1 more.
submit_work_sessionSubmit work for reviewAIdempotentInspect
End the authenticated worker's lease and submit typed result references for a separate review decision.
Inputs: session_id (required), result_summary (required), idempotency_key (required), used_steps, +2 more.
wait_for_project_activityWait for new 2ools project activityARead-onlyIdempotentInspect
Wait up to 20 seconds for attributable project events newer than a head_cursor or prior next_cursor. Returns events oldest-first so an outside agent can process them in order. If resync_required is true, paginate get_project_activity instead of advancing the cursor.
Inputs: project_id (required), after_cursor (required), limit, timeout_seconds.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 43 tools
Most tools have clearly distinct resource targets (project, build, version, artifact, agent, activity, feedback, work assignment). A few pairs like get_project and get_project_context could be confused, but descriptions clarify the differences.
Tool names consistently use snake_case verb-first patterns (get_, list_, create_, add_, remove_, approve_, reject_, cancel_, prepare_). Minor variation between create_ and add_ for similar actions, but overall the pattern is predictable.
43 tools is far beyond the typical well-scoped range. Even for a broad platform, this volume is overwhelming and likely makes it harder for agents to select the right tool. Many tools could be consolidated or grouped.
The tool surface covers the main lifecycle: project creation, building, versioning, sharing, artifacts, approvals, feedback, agents, activity, and work assignments. Minor gaps like lack of explicit delete/archive operations, which may be intentional given the immutable nature of versions and artifacts.
Maintenance
Related MCP Connectors
Remote MCP server exposing your Node Command fleet (files, shell, processes, browser) to AI hosts.
Remote MCP server for supportsheep: run AI interviews and manage support content for your blog.
Remote MCP server for supportsheep: run AI interviews and manage support content for your blog.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
Related MCP Servers
- AlicenseBqualityDmaintenanceProduction-grade MCP server that gives AI agents safe access to your local dev environment: filesystem, databases, processes, and OpenAPI specs.1540 npm4MIT
- AlicenseNot gradedqualityDmaintenanceMCP server for the OffByOne website factory, enabling agents to generate, validate, and manage full-stack web projects through safe, deterministic tools.MIT
- AlicenseNot gradedqualityFmaintenanceLocal stdio MCP server for working on Intern-hosted sites through guarded Git checkouts. It enables authentication, site preparation, testing, validation, and publishing via coding harnesses like Codex and Claude Code.37 npmMIT
- AlicenseAqualityBmaintenanceOutbound-only remote shell, detached long-running jobs, and temporary file courier for AI agents. Hosted Streamable HTTP plus stdio via npx -y @aicommander/mcp.113MIT