Skip to content

MCP Servers — Coding Agent Reference

MCP Servers — Coding Agent Reference

For the chat agent (Copilot in chat): Pull this file on-demand when writing problem statements for the coding agent.
Not intended for the coding agent — it doesn’t need to read this file.


🔑 Permission note

If a task requires a Cloudflare permission not listed below, ask @jaypatrick — additional permissions can be granted temporarily for any task. Don’t work around missing permissions; just ask.


Cloudflare MCP

The coding agent has access to a Cloudflare MCP server that can read and write Cloudflare resources directly via the CF API. Use this in problem statements instead of instructing the agent to run wrangler CLI commands for dashboard-side changes.

Account: 07a1f8d207654fe5d838174af4813126 (JK.com)

Current API token permissions

ScopeResourcePermission
AccountAI SearchEdit
AccountConnectivity DirectoryAdmin
AccountContainersEdit
AccountSecrets StoreEdit
AccountBrowser RunEdit
AccountWorkers PipelinesEdit
AccountAI GatewayEdit
AccountWorkers AIEdit
AccountQueuesEdit
AccountVectorizeEdit
AccountHyperdriveEdit
AccountCloudchamberEdit
AccountD1Edit
AccountWorkers R2 StorageEdit
AccountWorkers KV StorageEdit
AccountWorkers ScriptsEdit
AccountAccount SettingsRead
UserMembershipsRead
ZoneWorkers RoutesEdit
ZoneSSL and CertificatesEdit
AccountWorkers R2 StorageEdit

⚠️ The list above may be incomplete — more permissions may exist below the fold. @jaypatrick will provide the rest shortly; update this file when they do.

Known gaps (from prior inventory run)

The following were inaccessible during the March 2026 CF resource inventory — check if they’ve since been added above before assuming they’re missing:

  • Hyperdrive (now listed above — should be accessible)
  • AI Gateway (now listed above — should be accessible)
  • Vectorize (now listed above — should be accessible)
  • Logpush / Logs
  • Email Routing

What the CF MCP can do

Tell the coding agent to use the CF MCP server for:

  • KV namespaces — list, rename, create, delete, read/write keys
  • D1 databases — list, rename, run SQL queries
  • R2 buckets — list, rename, read/write objects
  • Workers Scripts — list, deploy, delete (rename = redeploy with new name)
  • Queues — list, create, delete, configure DLQ, manage subscriptions
  • Durable Objects — list namespaces and classes
  • Containers — list, create, delete applications
  • Workers Pipelines — list, create (no rename endpoint — requires delete + recreate)
  • Analytics Engine — list datasets
  • Hyperdrive — list, create, update connection strings
  • Vectorize — list, create indexes
  • Secrets Store — list, create, delete secrets
  • Workers Routes — list, create, delete
  • SSL/Certificates — list, manage

Known CF API limitations (cannot do via MCP)

  • Rename a deployed Worker script — CF has no rename endpoint. Renaming = deploy new name + delete old. Wrangler handles this on wrangler deploy when name changes in wrangler.toml.
  • Rename a Pipeline — no rename endpoint; must delete and recreate (data loss risk).
  • Rename a Workflow — workflows are identified by class name + script; rename requires new deployment.
  • Queue event subscriptions — must be managed in CF dashboard → Queues → Subscriptions tab (no API).

More servers

(Additional MCP servers to be documented here — @jaypatrick will provide details)