AI Agents, News & Updates

Anthropic Brings Back Third-Party Agents on Claude With Monthly SDK Credits

Anthropic has reversed its six-week-old ban on third-party agent tools against Claude subscriptions, but the restoration comes with a new metered credit system that ends unlimited agentic compute on flat-rate plans—effective June 15, 2026.

May 14, 2026 · 3 min read

Image by Anthropic

Anthropic Brings Back Third-Party Agents on Claude With Monthly SDK Credits

Anthropic announced today via @ClaudeDevs—its official developer communications account on X—that Claude subscribers will regain the ability to run third-party agentic tools such as OpenClaw and Conductor against their paid subscription plans. The announcement was made alongside the company's Claude for Small Business launch. The reversal takes effect June 15, 2026, and introduces a dedicated "Agent SDK credit" pool for each paid subscriber that is separate from interactive usage limits.

The move is a significant policy reversal. On April 4, 2026, Anthropic blocked all third-party agents from authenticating against Claude subscription plans after discovering that some subscribers were consuming compute worth hundreds of dollars per month under flat-rate plans priced at $20 to $200. The ban affected popular open-source tools, custom Agent SDK integrations, and CLI-based headless workflows—causing immediate backlash from the developer community.

The Six-Week Lockout and Why It Happened

The April 4 ban targeted what Anthropic described as unsustainable compute arbitrage. Unlike Anthropic's own products—Claude Code and Claude Cowork—which are engineered to maximize prompt cache hit rates and reduce per-token compute overhead, third-party tools were frequently unoptimized. A $20 Pro subscriber routing agent workloads through OpenClaw could generate token consumption equivalent to several hundred dollars at standard API rates, with Anthropic absorbing the difference.

The volume of inefficient third-party agentic traffic was large enough to create stability concerns across the platform, even as Anthropic expanded into the 300MW Colossus 1 data center with more than 220,000 GPUs. Demand for agentic workflows was outpacing sustainable supply at subsidized rates.

The New Agent SDK Credit System

The June 15 restoration comes with explicit constraints. Paid subscribers receive a fixed monthly "Agent SDK credit" that draws down at API token rates—not at the subsidized compute rate that interactive subscription limits reflect.

Credits are non-rollover: any unused balance expires at the end of the month. Once the credit is depleted, usage pauses until the next monthly reset—unless you have additional API usage credits enabled on your account, in which case those cover any overage at standard rates.

The following workflows draw from the new Agent SDK credit pool:

• OpenClaw and open-source agentic harnesses authenticating via Claude login
• Conductor (multi-agent IDE orchestrator) when using subscription authentication
• Third-party apps built on the Claude Agent SDK
• claude -p headless CLI invocations
• Claude Code GitHub Actions for automated CI/CD workflows

Claude Code in interactive mode and Claude Cowork continue to draw from the standard subscription usage pool and are not affected.

The End of Compute Arbitrage

The practical cost implications are significant. In early 2026, a $20 Pro plan could be leveraged through third-party agentic tools to run workloads with an effective API value of several hundred dollars per month—because Anthropic was subsidizing the difference between the flat subscription rate and actual token consumption.

That dynamic ends June 15. The $20 Agent SDK credit on a Pro plan reflects actual API token costs, not subsidized compute. Developers running high-volume autonomous pipelines through tools like OpenClaw or Conductor will need to audit their current usage now and decide whether to:

• Migrate to direct API keys, where usage-based billing has always applied and there is no credit cap
• Upgrade to a higher plan (Max 5x or Max 20x) for more monthly headroom
• Enable usage credits on their account so agentic workloads continue past the monthly credit limit at API rates
• Optimize agent efficiency to reduce token consumption before June 15

What Developers Should Do Before June 15

For developers using any tool that authenticates against Claude subscriptions programmatically:

1. Audit your agentic compute footprint now. Understand how many tokens your third-party tools consume per month under your current plan.
2. Check your tool's authentication options. Most Agent SDK-based tools support both subscription login and direct API key authentication—the latter is unaffected by the new credit cap.
3. Enable usage credits on your Anthropic account at platform.claude.com if you expect to exceed your Agent SDK credit allocation. This prevents pipelines from pausing mid-month.
4. Review Claude Code GitHub Actions workflows specifically—CI pipelines that invoke claude -p will now consume Agent SDK credits, not the general subscription pool.

Anthropic says the monthly credit becomes claimable starting June 15 and is available across all paid individual and team plans.

What Remains Unconfirmed

Anthropic has not disclosed the precise token rates at which Agent SDK credits are consumed—only that they draw down at "API rates." It is unclear whether prompt caching discounts apply to third-party tools using the new credit system the same way they do for Anthropic's own products.

The company has also not clarified whether the credit amount scales automatically if plan pricing changes in the future, or whether the allocation is a fixed dollar amount independent of plan cost adjustments. For developers planning multi-month production pipelines around this new credit pool, those details matter. Track the official documentation at code.claude.com/docs for updates.