Toolport
Comparison

Toolport vs Cloudflare MCP portals

Cloudflare brings Access-style identity and edge hosting for remote HTTP MCP. Toolport is local-first: one governed server set for the team, secrets on each machine, policy enforced in the gateway your agents already talk to. This page only claims what ships. Roadmap and non-goals are labeled.

14-day Team trial. No card.

How Toolport governance works

  1. Author in Teams. Admins set the shared server catalog, access, per-tool allow-lists, safety policy, rate limits, and webhooks.
  2. Sync to members. Each person's Toolport app pulls config. Toolport stores upstream credentials in each person's OS keychain and sends them to their configured upstream server when needed. Teams is designed to sync configuration and credential references, not upstream credential values. It strips recognized secret fields and rejects known secret patterns; do not paste secrets into shared commands, URLs or instructions.
  3. Enforce and report locally. The gateway applies policy on tool calls and reports self-reported apply receipts so the dashboard can show reported coverage, not only a config push.

This is cooperative enforcement, not a cloud proxy sitting on every call. A member who is not running Toolport is outside org policy. That is the trade for local-first, secretless control, and full support for stdio MCP servers.

Feature comparison

Cloudflare column reviewed 2026-10-06 against the linked Cloudflare MCP portal and Code Mode docs. Toolport column is what is live in the released app (policy, caps, webhooks, audit export, TOTP). Product capabilities change; check the sources below.

Topic Cloudflare MCP portals Toolport
Where it runs Remote HTTP MCP at the Cloudflare edge Local gateway on each machine (stdio + HTTP bridge), plus hosted or self-host Teams control plane
Secrets Platform / org patterns No org secret vault. Keys stay in each member's OS keychain
Shared server catalog Portal config Free and Pro include sync and five visible config versions; Team includes fifty.
Who can use which server Portal access controls Per-server member and group grants (Team+ to edit)
Per-tool allow / deny Portal tool policy Org allow-lists and deny-lists enforced in the local gateway (Team+)
Safety policy Platform controls Destructive block, human approval, quarantine, content defense; org can only tighten
Tool execution mode Code Mode with off, opt-in, default-on, or enforced policy Server-side Code Mode through toolport_run_script, off by default; opt-in under Advanced settings, or force it on with TOOLPORT_CODE_MODE=1. Control is global, not per client
Policy coverage reported Inline path = enforced at edge Self-reported apply receipts from each member's app, with timestamps; not device attestation
Runaway tool-call protection Spend / identity budgets (model and platform) Hard tool-call rate limits (day/month, member/tool axes) in the gateway (Team+); 80% webhook warn
Spend budgets Dollar limits with block at model layer Monthly estimated-cost budget with admin email alert (Team+). Not a model-meter hard block
Audit trail Analytics + Logpush-style export Usage and events CSV; Free 14-day floor; opt-in per-call export (tool, time, duration, ok, args hash; never args or results)
Webhooks / alerts Platform notifications HMAC-signed webhooks, retries, all or security filter; Free includes one security channel
Dashboard sign-in Cloudflare Access (any IdP, MFA, device posture) GitHub / Google OAuth, magic link, and TOTP 2FA
SSO (SAML / OIDC) + SCIM Yes via Access / IdP Roadmap (Enterprise). Contact for scoping Roadmap
Device posture Yes (Cloudflare One) Not a product lane (local-first) N/A by design
Self-host control plane Different product shape Yes: Docker image on your network

Sources: Cloudflare MCP portals and Cloudflare Code Mode changelog, both reviewed 2026-10-06.

Where Cloudflare is the better fit

  • Any-IdP Access, MFA posture, and enterprise identity as the front door
  • Remote HTTP MCP at a global edge, with no local stdio server requirement
  • Device posture and Cloudflare One style zero-trust stack
  • Model-layer spend limits and Logpush-class streaming (AI Gateway / platform)

Where Toolport is stronger

  • Local and stdio MCP, multi-agent desktop setups
  • Configuration and credential references, with recognized secret fields stripped
  • Tool-layer rate limits and tighten-only safety on the gateway
  • Apply receipts so cooperative policy has reported coverage
  • Self-host the Teams control plane on your own infrastructure

Straight answers

Is Toolport a security boundary like Cloudflare Access?

No. Org policy is authored in Teams and enforced by each member's local Toolport gateway. A machine that is offline or not running Toolport is outside that policy. Apply receipts are self-reported coverage records from each member's Toolport app, with timestamps. They show what a client reported as applied. They are not device attestation or proof that a client that bypassed Toolport followed policy.

Where do API keys live?

Toolport stores upstream credentials in each person's OS keychain and sends them to their configured upstream server when needed. Teams is designed to sync configuration and credential references, not upstream credential values. It strips recognized secret fields and rejects known secret patterns; do not paste secrets into shared commands, URLs or instructions.

Do you hard-block runaway agents?

Yes for tool-call rate limits on Team: day and month caps by team, member, group, or tool, enforced in the gateway. Monthly spend budgets alert admins on estimated MCP tool spend; they do not block cloud model billing.

What about SSO and SCIM?

Dashboard sign-in today is OAuth, magic link, and TOTP 2FA. SAML/OIDC and SCIM are Enterprise roadmap, scoped with you before you commit, not self-serve checkboxes today.

What about API tokens for CI?

On Team, you can mint a read-only API token that pulls the full shared server config for deploy or CI. It is not per-server or otherwise scoped today.

Try the governance path that keeps keys local

Team is $19 a month flat for up to 10 people, adding access control, per-tool policy, rate limits, signed webhooks, unlimited retained audit history, and more.

14-day Team trial. No card.

Related: Security model · Pricing · Source