Read
walling.read Search and read walls, sections, bricks, collections, and proposals. Always granted. Without it the connection is read-only.
Docs · MCP server
Walling MCP server
Connect Walling to any AI client that speaks the Model Context Protocol — Claude, Cursor, ChatGPT, and more. Bring your workspace context in, capture to your inbox, and propose canvas changes you approve in the app.
Endpoint
Point your AI client at the Walling MCP server:
https://mcp.walling.io/mcpThe server uses standard OAuth. The first time you add it the client opens a Walling consent screen where you sign in, name the connection, and pick which permissions to grant. After that the client manages tokens automatically — there are no API keys to copy.
Connect your AI client
Claude. In Claude’s connectors settings, add a new connector and paste the URL above. Walling appears with our logo in the connector list once authorized.
Cursor and ChatGPT. In Cursor’s settings or ChatGPT’s connector picker, add a new MCP server with the same URL. Both clients render the Walling logo and link to walling.io directly from their server picker.
Other MCP clients. Any MCP-compatible client works the same way — point it at the endpoint and complete the OAuth flow. The server advertises its capabilities (tools, resources, and the Walling Skill prompt) on connect, so no extra configuration is needed.
Permissions and write policy
When you authorize a connection, you choose which scopes to grant. Scopes map directly to which tools the agent can use.
Write — propose
walling.write Let the agent propose canvas changes (new walls, sections, bricks; edits and deletes). Changes appear as pending proposals in the Walling app for you to approve or reject.
Write — execute
walling.execute Same writes as above, but applied directly without a review step. Grant this only to clients you trust to act on your behalf.
Manage
walling.manage Create and configure scheduled AI agents and email pipelines on your account.
Quick capture to your inbox is always available with the Write scope and never goes through a proposal — it lands as a brick in your inbox immediately, so you can use the agent as a fast capture tool.
What the agent sees in your workspace
The agent never browses arbitrary Walling content. It only sees what your scopes allow inside the workspace tied to your account, scoped further by the tool it calls (a collection, wall, section, or brick id). Search results, bricks, and proposals returned to the agent only contain content the connection’s scopes authorize.
Every Walling tool advertises a friendly title and standard MCP safety hints to the client. Destructive actions — cancelling a proposal or deleting an email pipeline — are marked so clients can ask for confirmation before calling them.
Available tools
Reads (always available)
- Search Walling workspace
- Read wall · Read section · Read brick
- List collections · List walls · List section view types
- Get session context — the agent’s current orientation in your workspace
Writes (with walling.write or walling.execute)
- Save to Inbox — quick-capture a note as a brick in your Inbox
- Plan canvas changes — create or edit walls, sections, and bricks in one batched, reviewable operation
- Set section view type · Update section view settings · Set brick position(s)
Proposals (always available)
- List canvas proposals · Read canvas proposal · Cancel canvas proposal
Management (with walling.manage)
- AI agents — list, read, create, update, enable/disable, run now
- Email pipelines — list, read, create, update, pause, resume, delete
Reviewing proposals
When a connection runs in propose mode, every canvas write becomes a proposal. The Walling app shows a notification with a link to the proposal where you see the full plan — what walls, sections, and bricks will be created or changed — and choose Approve or Reject. Proposals you don’t act on expire automatically. Approval happens only in the Walling UI; there is no “approve” tool.
Manage your connections
Open Settings → Connections in Walling to see every active MCP connection, what scopes it has, and when it was last used. You can revoke any connection from here at any time. Revocation takes effect immediately — the agent loses access on its next call.
FAQ
Common questions
Do I need a paid plan to use the MCP server?
The MCP server is available on all plans, but write scopes
(walling.write,
walling.execute) and management
(walling.manage) may require an eligible
plan. Check Settings → Billing for
what your current plan includes.
Does the agent see content from other workspaces I belong to?
No. Each connection is tied to one workspace. To use Walling MCP in another workspace, switch to that workspace in the Walling app and authorize a separate connection from your AI client.
Can the agent permanently delete my content?
No. The destructive tools available over MCP are limited to cancelling pending proposals and deleting email pipelines (both confirmed actions). Deleting bricks, sections, or walls happens only through canvas_plan — and on connections in propose mode that means a proposal you review before anything is removed.
What happens to my data when I revoke a connection?
The agent loses access on its next call. Walling does not delete content the agent created, but no further reads or writes from that client are accepted. Existing proposals remain in your queue until you approve, reject, or let them expire.
Where can I report a problem with the MCP server?
Use the ? menu in the Walling app to open a support ticket, or email support@walling.io with the name of the connection and the time of the issue. For privacy details, see our privacy policy.