What did Hlido score Recurse?

Recurse scored 73/100 (STEADY) in Hlido's independent, hands-on review.

How much does Recurse cost?

Hlido recorded the pricing model as: Free tier. Derived from evidence Hlido holds, not vendor-supplied; verify current prices on the vendor's pricing page.

Does any vendor pay Hlido for placement?

No. Hlido takes no money from the agents it rates — scoring weights stay private and the evidence behind every verdict is public.

Recurse

Name: Recurse review
Item: Recurse
Rating: 73
Author: Hlido Editor

Infrastructure · tested 2026-10-02 · re-test due 2027-01-02 · by the Hlido desk, not the vendor

In short: Hosting-for-agent-apps that treats MCP as the output format — build an agent app, deploy it with one CLI command, publish it as an MCP any client can call. Exactly the shape the agent-to-agent era needs; depth still to be proven hands-on.

5 PASS · 0 FAIL of 5 public-surface claims

Visit Recurse → scorecard.json

Quick answer

Recurse scores 73/100 (STEADY) on Hlido’s independent, hands-on test (reviewed 2026-10-02). STEADY (73) because the public surface is concrete and well-aimed for agent builders: a real CLI and SDK, MCP as the deployment target, worked examples, and self-serve docs/pricing — strong agent-to-agent fit with low ob Pricing: Free tier.

Hlido Editor · 2026-10-02

Recurse's premise is narrow and well-aimed: you build an agent app, deploy it, and it becomes an MCP server that Claude, Codex or any MCP client can use — with Recurse handling the hosting. The developer path is concrete on the public surface: install the SDK (pip install recurse-sdk), deploy with 'recurse deploy . --as mcp', or start from an example in the gallery. There is even a copy-paste SKILL.md prompt so a coding agent can set the project up for you, which is a thoughtful nod to how people actually scaffold these days. For Hlido, the standout is positioning: most tools make you integrate an agent into something; Recurse makes the agent app itself a callable MCP primitive, which is the right abstraction for an ecosystem where other agents are the consumers. The surface advertises a gallery, docs, pricing and a blog — more self-serve substance than most entrants at this stage. Where Hlido stays cautious: everything here is the happy path as marketed. Hlido did not deploy an app, call the resulting MCP, or test reliability, cold-start behaviour, limits or what the hosting actually guarantees — and 'we handle the hosting' is a promise whose value is entirely in the operational detail a landing page cannot show. A credible, correctly-shaped piece of agent infrastructure reviewed from its public surface only.

Why STEADY

STEADY (73) because the public surface is concrete and well-aimed for agent builders: a real CLI and SDK, MCP as the deployment target, worked examples, and self-serve docs/pricing — strong agent-to-agent fit with low obvious trust friction. Not higher because none of it is exercised: hosting guarantees, reliability, limits and the real deploy experience are unverified from the outside, and confidence is low on a single-surface read.

Public-surface checklist

PASS Homepage loads (required)
PASS Primary value prop (required) — 'Build and deploy agent apps' / publish as MCP
PASS Cta present (required) — 'Get started' / 'Use the CLI'
PASS Pricing or access — Pricing page linked in nav (concrete tiers not captured on homepage)
PASS Evidence or demo — Examples gallery + docs + copy-paste SKILL.md on the surface

What it does well

Clear one-command deploy story: 'recurse deploy . --as mcp' turns an agent app into a callable MCP server
Publishes agent apps as MCP servers any client (Claude, Codex, generic MCP) can consume — the right primitive for agent-to-agent use
Real self-serve developer path: Python SDK (pip install recurse-sdk), examples gallery, docs
Ships a copy-paste SKILL.md so a coding agent can scaffold the project
Managed hosting removes the ops burden of running an MCP endpoint

What it fails at

Hosting guarantees, uptime, cold-start and scaling behaviour are unstated on the public surface
No visible detail on limits, data handling or security posture of hosted agent apps
Deploy experience and MCP reliability were not exercised by Hlido — happy-path marketing only
Pricing is linked but concrete tiers were not captured on the reviewed surface

Best for

Developers who want to ship an agent app as an MCP server without managing infrastructure
Teams standardising on MCP as the interface between their tools and Claude/Codex clients
Builders who prototype with a coding agent and want a one-command path to a hosted endpoint

Not recommended for

Teams that require self-hosting or full control over the runtime and data path
Production workloads needing published SLAs/limits before committing (not visible on the surface)
Anyone who needs non-MCP delivery as the primary interface

Pricing & access

ModelFree tier
Pricing findable on the public surfaceFAIL A pricing page is linked in the nav, but concrete tiers were not visible on the captured surface (tested 2026-10-02)

Derived from Hlido-held evidence only (public-surface capture + editorial text); not vendor-supplied; re-derived daily. Verify current pricing on the vendor's pricing page. Last verified 2026-10-02.

Agent relevance

API CLI MCP SDK Behavioral-testable

First-class: build an agent app with the Recurse SDK, deploy it via the CLI, and it is published as an MCP server any MCP client can call. The deployed artifact is itself an agent-consumable interface, which is the strongest possible agent-integration posture.

Agent-friendly score: 8/10

Evidence

Deploy an agent app as an MCP server usable from Claude/Codex/any MCP client — source (2026-10-02) verified
Python SDK and CLI exist (pip install recurse-sdk) — source (2026-10-02) verified
Examples gallery and docs are available — source (2026-10-02) verified
Managed hosting reliability / guarantees — source (2026-10-02)

scorecard.json · registry · methodology

More: compare agents · best of · developer tools · incident registry

Verdict by Hlido Editor, our automated editorial system · Method: public-surface-tier-2+editorial-narrative-v2 · Methodology version 2026.09 · Next review due 2027-01-02

How this page was produced. The scores, claim verdicts and evidence come from automated hands-on testing of the product’s public surface. The written analysis is drafted by an AI system, and pages publish without a person reviewing each one. Hlido publishes this record and answers for it — tell us if anything here is wrong and we will correct it.

Embed this trust badge

Live, always-current independent score — free to embed on your site or README. No vendor pays for placement.

Markdown

[![Hlido trust score](https://hlido.eu/badge/recurse.svg)](https://hlido.eu/check/?agent=recurse)

HTML

<a href="https://hlido.eu/check/?agent=recurse"><img src="https://hlido.eu/badge/recurse.svg" alt="Hlido trust score"></a>