Our take: Install Surfil on a Mac or Linux dev machine if you want measured per-request token costs and secret-leak blocking for Claude Code, Cursor or Codex without your source code leaving the device.
Main limitation: No free tier (Starter is $5/mo), it needs a Mac or Linux terminal, and Guard enforcement, Mind and Radar/Bench require Pro or higher.
By the AI Tool Atlas editors.
Based on the vendor's official documentation; we have not tested this product hands-on. What “documented overview” means
Is Surfil right for you?
Install Surfil on a Mac or Linux dev machine if you want measured per-request token costs and secret-leak blocking for Claude Code, Cursor or Codex without your source code leaving the device. Based on the vendor's documentation, not on a hands-on test.
Best for: Developers running several AI coding agents
Also a good fit for
Teams needing cost visibility on agent token spend
Security-conscious engineers using MCP tools
Consider another tool if no free tier; entry Starter plan is $5/mo
Or if
Requires a Mac or Linux machine; Windows not mentioned
Guard enforcement, Mind and Radar/Bench require Pro or higher
Surfil pricing and usage limits
Prices checked Oct 1, 2026 on the official US pricing page. Official pricing ↗ Prices and limits change often.
Vendor pricing page re-checked automatically 11 days ago: no changes detected.
The official Surfil pricing page as we saw it. surfil.com, captured Oct 1, 2026
Free access and trial conditions
No free tier.
14 days of Pro for $1, then $25 per month unless cancelled; one trial per account and the $1 is not refunded.
For: Organizations needing SSO and policy controls
Limits
Practical-unlimited credits
Includes
Proof, Fleet Control, Charter, Policy
SSO (OIDC) and SLA
Usage caps, credits and extra costs
Agent execution is unmetered; each server-signed output (receipt, benchmark, attestation) costs one Credit, and top-ups are available.
Not stated.
What is Surfil, and how does it work?
Surfil is a locally installed control plane that sits between AI coding agents such as Claude Code, Cursor, Codex, Copilot and MCP-based agents and their model providers. A one-command install on Mac or Linux routes agent requests through an on-device pipeline covering security, cost, quality, observability and memory, and consolidates scattered MCP server configs into one endpoint.
Its modules include Cap (per-request cost ledger from provider token counts), Guard (blocking API keys, tokens or injected instructions before they leave the machine) and Mind (shared memory across agents with staleness flags). Savings and benchmarks are issued as Ed25519-signed receipts that can be verified offline. Agent traffic is unmetered; each signed output consumes one Credit.
Surfil's official site: how the product presents itself. surfil.com, captured Oct 1, 2026
What you provide and what you get back
What you give Surfil and what comes back, from its documentation
You provide
AI agent requests intercepted at the agent base URL
MCP server configs
Memory facts written by agents
You get back
Per-request token cost ledger
Signed Ed25519 receipts verifiable offline
Blocked or flagged risky requests
Where you can use it
Platforms
Desktop, API
Account
Required
Platforms
Mac and Linux terminal install
Supported agents
Claude Code, Cursor, Codex, Copilot, MCP-based agents
Modules
Cap, Guard, Mind, Radar, Bench, Pilot
Proofs
Ed25519-signed receipts, verifiable offline
Annual discount
20%
Best use cases for Surfil
Scenarios are worked out by our editors from the documentation; the actual result depends on your input and plan.
Seeing what Claude Code, Cursor and other agents actually spend per request
Who
Developers running Claude Code or Cursor
You provide
One-command install and normal agent use
Expected result
A per-request ledger of token counts and cost from provider responses
Source code never leaves the machine; only authentication, metering and signing happen at the edge You keep code private since only metadata reaches Surfil serversdocumentedsource (opens in a new tab)
14-day Pro trial for $1 with 250 signed outputs included You can measure real savings on your own traffic for $1 before committingdocumentedsource (opens in a new tab)
Agent execution is unmetered; only signed outputs cost credits Running agents heavily does not raise your billdocumentedsource (opens in a new tab)
No free tier; entry Starter plan is $5/mo Budget-minded hobbyists cannot try it for freedocumentedsource (opens in a new tab)
Requires a Mac or Linux machine; Windows not mentioned Windows developers cannot install itdocumentedsource (opens in a new tab)
Guard enforcement, Mind and Radar/Bench require Pro or higher Teams needing enforcement or memory cannot stay on the cheapest tierdocumentedsource (opens in a new tab)
Team plan is set up through sales rather than self-serve Team buyers cannot self-serve checkout and must wait for salesdocumentedsource (opens in a new tab)
How to use Surfil for measuring and securing AI coding agent traffic
Set up Surfil
Start the $1 Pro trial or pick a plan on the pricing page
Run the one-command install from the quickstart in a Mac or Linux terminal
Let the installer point your agent base URL at the local runtime
Run your normal Claude Code or Cursor session
Example: measuring and securing AI coding agent traffic
Use your agent as usual for a coding task
Open the Cap ledger to read tokens and cost per request
Set Guard to enforce and try sending a fake API key in a prompt
Generate a savings receipt and run surfil verify on it
Review and export the result
Confirm surfil verify reports the receipt as valid offline
Confirm Guard blocked the test secret and that uninstall would restore your original configs
Surfil integrations and API
Plan comparison on the official pricing page. surfil.com, captured Oct 1, 2026
Native integrations and export formats
Export: .surfilpkg
API access and separate billing
Surfil has no public API in its official documentation. Public edge API at api.surfil.com documented in the reference API documentation ↗
Privacy, data use and commercial rights
Summarized from the vendor's current policies; this is not legal advice and a vendor's badge is not an independent audit.
How your inputs and outputs are handled
Source code stays on the device; telemetry is metadata only (counts, durations, model names, token totals).
Not stated beyond that prompts and source content are not collected.
Operational metadata kept for billing and audit; ciphertext memory kept until you delete it or close the account, then purged.
One server-signed value-output such as a savings receipt, benchmark result or attestation; agent execution itself is unmetered. source (opens in a new tab)
What happens when my Surfil credits run out?
Execution keeps working, but new signed outputs wait until window or monthly credits refresh or you top up. source (opens in a new tab)
How does the Surfil $1 trial work?
You get 14 days of Pro for $1 including 250 signed outputs, then $25 per month unless you cancel; one trial per account and the $1 is not refunded. source (opens in a new tab)
Does Surfil send my source code to its servers?
No. The privacy policy says source code never leaves your device; the edge only authenticates, meters and signs, and telemetry is metadata only. source (opens in a new tab)
Does downgrading Surfil delete my data?
No. Weave, receipts and memory survive a downgrade; paid features just stop producing new signed outputs. source (opens in a new tab)
Not verified: Windows support; Training-use statement beyond metadata-only telemetry; Credit amounts per plan; Privacy page says it is illustrative, not final legal wording.
How we documented and maintain this page
Every fact about Surfil above comes from the 4 official pages listed here, last compared on Oct 1, 2026; we have not run Surfil ourselves, so it has no editorial score. How we check and re-check pages.
Article updated Oct 10, 2026; prices checked Oct 1, 2026. Outbound links to vendors are plain links; we do not earn commissions. How we review · Report an error