Recurse
recurse.run- Category
- AI Agents
- Rank
- No. 2341Tools index
Previous survey · No. 2298 ·
- Pricing
- Freemium
- Type
- TOOL
- Use case
- Agent Building
- Interfaces
- MCP
- Date
About
Recurse is a serverless harness that lets a coding agent design, test, and deploy custom "specialist" sub-agents defined by a YAML manifest (agent.yaml), which pins the specialist's inputs, outputs, and the exact tools it may call. Deployed specialists can be exposed as tools, MCP servers, or bots, and the harness runs many candidate variations of a specialist, refining them against a measurable bar until one passes (the site's write-ups cite examples like a level designer and an RNA-sequence design agent).
What it does
Recurse's site presents it as a toolbox for coding agents: one account that reaches many third-party data APIs, plus a searchable library of task recipes, mostly SEO, sales, company research and social-media analysis, which an agent can build into deployable tools.
Stated on the product site
- Access model
- Many provider APIs reached through a single Recurse account
- Scope
- Building and deploying tools and agents that call those APIs
- Recipe language
- Recipes are Python and route each provider call through one shared call function
- Testing labels
- Each recipe is labelled with how far it has been tested, and most carry a not-live-tested label
- CRM integration
- HubSpot recipes run against the user's own HubSpot account
- Agent workflow
- The page shows the recipes being used from Codex on a local machine
Not stated on the site
- What using the APIs costs after the sign-up credit the page mentions.
- How many recipes have been run end to end against live providers, since most are marked as specifications.
- Where deployed tools run and how they are exposed to an agent once built.
- How the specialist sub-agent workflow in the catalogue entry relates to the API toolbox this page presents.
Written from the product site at recurse.run.
Tags
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.