The contextual backbone for your engineering team and AI agents.
Manage features, solution designs, tests and architecture decisions in one place, then hook that knowledge directly into coding agents and IDEs so everything gets built from a place of full contextual awareness.
AI agents work better with real business context
Without project knowledge, agents guess. With Specsource, they get it right first time.
Describe how you want to build it, so engineers and agents are aligned
Describe how you want to build it. AI reads your specs and proposes changes. You accept or reject each one with full traceability.
AI-powered synthesis
Feed in API docs, meeting notes, Figma designs, PDFs. Specsource synthesises them into structured specifications for your approval.
Write tests your team can read and agents can generate
AI agents read your features and designs, then generate test cases automatically. Run them step by step and share interactive reports with clients.
Set the rules once — for your team and every agent
Stop repeating context in every prompt or onboarding doc.
Define coding conventions, architecture rules and project background once. Agents fetch the full picture via MCP so they build from awareness, not assumptions.
# Auth Rules - User IDs are text (Clerk), not UUID - All RLS policies use auth.jwt() ->> 'sub' - Every server action checks permissions # Naming - Files: kebab-case - CSS modules: camelCase classes - Imports: s from './component.module.css'
Generate a ready-to-paste prompt from your project instructions, or let agents pull them automatically via MCP.
You are working in a Next.js 16 project with Clerk auth and Supabase. User IDs are text strings (Clerk format), never UUIDs. All RLS policies must use auth.jwt() ->> 'sub'. Follow kebab-case file naming and camelCase CSS module classes.
One source of truth for distributed teams
Different timezones, different locations, same specification. Specsource keeps every contributor aligned.
Let clients ask questions about the project
Share a portal link with stakeholders. They ask questions in plain English and get answers sourced from your specs, decisions and tests. No login required.
Ask questions about the project in plain English. Results are sourced from specs, decisions, features, and tests.
JWT was chosen over server sessions for stateless horizontal scaling. The decision was driven by the need to scale auth across multiple regions without shared session state. TOTP-based MFA was selected over SMS due to SIM-swap security risks.
Your IDE reads your specs via MCP
Connect Specsource to Cursor, Claude Code or any MCP-compatible tool. Your agent fetches features, decisions and test cases before writing code, so it builds what you designed.
get_feature_contextGet everything an agent needs to build a feature — spec, linked designs, ADRs, tests, and instructions in one call.
search_entitiesSearch across all entity types — features, tests, designs, and decisions — by keyword.
get_testRetrieve a test by ref with all steps, expected results, and linked features.
get_project_contextEverything in one call — instructions, features, decisions, systems, tests, and business context.
Features
Define what your product does with acceptance criteria, user stories, and goals.
Architecture Decisions
Record the why behind technical choices. Context, alternatives, and consequences.
Tests & Reports
Structured test cases with steps and expected results. Share reports with clients.
Agent Instructions
Per-project AI coding guidelines that carry across all agent implementation threads.
Systems & Data Flows
Map your architecture. Define services, databases, and how data moves between them.
Tags & Relationships
Cross-link any entity. Tests verify features, features implement designs.