Ref Tools

What Ref Tools does

Ref Tools, now presented in the vendor documentation as Ref, combines two products for software teams. Ref Context gives MCP-compatible coding agents access to public and private technical documentation. Ref Plans provides a shared plan editor and orchestration layer that can send structured work to coding agents. This is a more specific proposition than a general AI coding assistant: Ref is intended to improve the context and coordination around an agent, not replace the agent or IDE.

Ref Context exposes documentation-oriented MCP tools, including search and page retrieval, so an agent can check current API behavior before producing code. It can also index private GitHub repositories, PDFs, and Markdown resources. The installation documentation lists Claude Code, Codex CLI, Gemini CLI, VS Code, Zed, Cursor, Windsurf, Kiro, Goose, OpenCode, Cline, Roo Code, Devin, ChatGPT, and other clients. Ref Plans adds shareable plans, review requests, change control, team visibility, rich content, and multi-agent orchestration. Integrations listed by the vendor include Cursor background agents, Devin, Warp, Claude Code Web, Codex, Linear, Asana, GitHub issues and pull requests, and Slack.

Verified pricing

The official documentation publishes four current tiers, with credits shared by Ref Context and Ref Plans. Free costs $0 and includes 200 one-time credits, unlimited plans, three small repositories, one large repository, and 100 PDF pages. Pro costs $50 per month and includes 6,000 monthly credits, unlimited plans, 50 small repositories, five large repositories, and 5,000 PDF pages. Max costs $200 per month and includes 30,000 monthly credits, unlimited small repositories, 25 large repositories, and 25,000 PDF pages. Enterprise uses custom pricing and limits, with SSO and dedicated support.

Pay-as-you-go usage is available on Pro and above at $10 per additional 1,000 credits, charged after the billing cycle. The documentation says legacy $9 per month Basic and Team subscriptions continue with 1,000 credits and that older $19 Basic subscriptions also continue, but neither is offered to new buyers. These figures are much more useful than a generic “freemium” label. This file remains flagged for manual verification because the main ref.tools homepage and conventional pricing route exposed little readable content in the scheduled fetch, while the exact prices came from docs.ref.tools.

Strengths, tradeoffs, and alternatives

Ref's main advantage is that documentation search and planning share one credit pool and one agent-oriented workflow. A developer can ask an agent to verify a library call, attach private repository context, write a reviewed implementation plan, and launch work without treating each stage as a disconnected tool. MCP support also avoids a proprietary client dependency. Read the Model Context Protocol overview and the MCP builder guide before granting access to private resources.

There are meaningful tradeoffs. Credits are consumed by both documentation queries and AI generation in the plan editor, so heavy multi-agent use can make monthly consumption less predictable. The free allowance is one-time, not 200 recurring credits each month. Private indexing also introduces security and freshness questions: buyers should verify repository permissions, deletion behavior, sync timing, retention, and whether generated plans expose sensitive code or issue details. Ref helps agents find source material, but it cannot guarantee that every answer is correct or that every indexed page is current.

Ref complements rather than replaces Cursor, GitHub Copilot, or Claude. Compare the total workflow, not just the subscription price. If a team already has reliable repository search and keeps plans in pull requests, Ref may duplicate existing processes. If agents repeatedly hallucinate APIs, lose architectural decisions between sessions, or require humans to copy context among tools, the combination can save time.

Practical evaluation

Run a two-week pilot on one active repository. Index only the documentation and code needed for that project, then create 20 representative tasks: API migrations, unfamiliar framework questions, bug investigations, and features requiring a reviewed plan. Measure source relevance, unsupported claims, time to an accepted plan, implementation rework, credits per completed task, and the number of times a developer must leave the coding client to search manually.

Start with Free for a functional check, but 200 one-time credits may not support a statistically useful team test. For Pro, divide the $50 monthly fee by completed tasks and include any $10-per-1,000-credit overage. Test revocation by removing a repository and confirming it disappears from search. Ref is strongest for developers and teams using multiple MCP-capable agents who need current documentation and a shared planning layer. It is less compelling for occasional coding assistance, teams unwilling to index private material, or workflows already standardized around repository-native documents.