Hoplite - Cloud Coding Agents That Ship Pull Requests
Hoplite deploys cloud coding agents that read your repo, run tests, drive browsers, and open GitHub pull requests. Built on Y Combinator S26.
TL;DR
TL;DR: Hoplite is a cloud-based coding agent platform that connects to your GitHub repo, executes tasks in isolated environments, verifies changes with tests, and opens pull requests — with MCP and CLI interfaces for integration.
Source and Accuracy Notes
⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.
- Project page: hoplite.sh
- YC company page: ycombinator.com/companies/hoplite — confirms Y Combinator S26 batch membership
- Documentation: hoplite.sh/docs
- MCP server: hoplite.sh/docs/mcp-server
- License: proprietary SaaS (cloud service, no open-source license)
What Is Hoplite?
Hoplite is a cloud coding agent platform that automates the path from a software task to a reviewed pull request. Each agent runs in an isolated cloud development environment, where it can read repository files, edit code, run test suites, interact with browser previews, and open GitHub pull requests — all while you watch and approve sensitive actions.
The platform is designed for teams that want agent assistance without self-hosting infrastructure. Agents work in persistent threads, so you can iterate on feedback without re-explaining context.
How It Works
Step 1: Connect a Repository
Link a GitHub repository through the Hoplite GitHub App. Configure the base branch, a setup script for the sandbox environment, and any encrypted environment variables the agent needs.
Step 2: Start a Thread
Initiate a thread with a task description — for example, “add rate limiting to the /api/users endpoint” or “migrate the auth module to use JWTs.” The agent picks up the context from the repository and begins working.
Step 3: Follow and Approve
Watch the agent’s progress in the web UI. The agent can request approval for sensitive actions — pushing to protected branches, publishing releases, or accessing encrypted secrets — before proceeding.
Step 4: Receive a Pull Request
The agent opens a GitHub pull request with its changes, including test output and verification evidence. Continue iterating in the same thread as you review; the agent retains context from earlier turns.
Interface Options
MCP Server
Hoplite exposes an MCP server at https://api.hoplite.sh/mcp using streamable HTTP with OAuth protection. You can drive Hoplite from any MCP-compatible client, giving agents access to Hoplite’s task execution alongside other tools.
CLI
The CLI (hopl) supports importing local Claude Code, Codex, or OpenCode conversation history and managing project-specific skills. Install via the CLI documentation.
LLM Index and Agent Capabilities
Two lightweight discovery endpoints help other agents understand Hoplite’s surface:
- hoplite.sh/llms.txt — compact site index oriented toward agent consumption
- hoplite.sh/agents.txt — MCP capability and authorization directory
Security Model
Hoplite enforces several constraints that are worth knowing before connecting a private repository:
- Repository access is scoped through a GitHub App installation — not personal access tokens
- The web app and MCP server both require an authenticated Hoplite account and workspace-level authorization
- Encrypted environment variables are decrypted only inside isolated sandbox environments
- Credentials, access tokens, and private repository content are never sent to public discovery endpoints
Pricing
Hoplite uses a sandbox-capacity model. Plans are billed by the number of concurrent sandbox environments available to your workspace. Model usage is metered against a workspace credit balance separately from sandbox capacity. See the pricing page for current tiers.
FAQ
Q: Does Hoplite run locally or in the cloud? A: Hoplite runs entirely in Hoplite’s cloud infrastructure. There is no self-hosted option.
Q: What LLM models does Hoplite use? A: Hoplite is model-agnostic. The platform routes tasks to models based on context and pricing. Exact model selection is managed by Hoplite internally.
Q: Can I use Hoplite without connecting my entire organization? A: Yes. Individual developers can connect personal GitHub repositories through the GitHub App without requiring org-level approval.
Q: How does approval-based execution work? A: You configure which actions require explicit approval (e.g., pushing to main, accessing certain env vars). The agent pauses and prompts you in the UI before proceeding with those actions.
Q: Is there an open-source version? A: No. Hoplite is a proprietary SaaS product. There is no open-source or self-hosted edition.
Practical Evaluation Checklist
- Connect a test repository with a clear, bounded task
- Verify the agent reads the correct files and respects
.gitignore - Check that test failures in the sandbox are surfaced in the PR summary
- Test the approval flow by triggering an action that requires consent
- Inspect the GitHub App permissions requested — confirm only what is needed is granted
- Evaluate response time and context retention across multi-turn threads
Security Notes
- Review the GitHub App permissions before installing — scope is set per repository or organization
- Encrypted env vars are injected at sandbox startup, not at build time
- The MCP endpoint uses OAuth; do not expose the credential in logs or shared configs
- Hoplite’s own docs note that you should never send production credentials to public discovery endpoints
Conclusion
Hoplite targets teams that want coding agent automation without managing the underlying infrastructure. Its isolation model, approval-gated execution, and GitHub-native PR workflow make it a practical middle ground between a raw LLM API and a fully self-hosted agent setup. The MCP interface extends its reach to any MCP-compatible client, and the CLI import feature lowers the barrier for teams already using Claude Code or Codex.
For a cloud coding agent that ships actual pull requests with verification evidence, Hoplite is worth evaluating — particularly if your team wants the output to be review-ready rather than a raw diff that still needs significant cleanup.
hoplite.sh — backed by Y Combinator S26 per ycombinator.com/companies/hoplite
Related Posts
ai-setup
Recall – Persistent Memory for Claude Code via MCP Hooks
Recall gives Claude Code a permanent memory store that survives session restarts and context compaction. Four hooks capture and restore context automatically — with cloud SaaS or self-hosted options.
2/28/2026
dev-tools
Automotive Skills Suite for AI Engineering
Evaluate Automotive Skills Suite for APQP, ASPICE, HARA, safety-plan, and DIA workflows with setup notes, governance risks, and SME review guidance.
5/28/2026
dev-tools
awesome-agentic-ai-zh Roadmap Guide
Explore awesome-agentic-ai-zh as a Chinese agentic AI learning roadmap, with setup notes, track selection, study workflow, and evaluation guidance.
5/28/2026