continuum

Overview

Continuum is an MCP server that provides shared memory and orchestration for multiple coding agents. It creates a centralized knowledge base using SQLite with vector embeddings, allowing different AI clients like Claude Code, Codex, and Cline to share learned experiences and avoid redundant explanations. The server implements a canonical workflow with a 4-phase dispatch protocol (Intake → Research → Verify → Handoff) and maintains a registry of agents to prevent conflicts across git worktrees.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• project_create

  Create a new Continuum project for a repository path

  Try:Create a Continuum project for /absolute/path/to/my/repo
• plot

  Read the PLOT.md orchestration protocol for a project

  Try:Read the PLOT.md for my projectTry:Show me the orchestration protocol
• registry_list

  List active agents with their metadata and reserved paths

  Try:Show me all active agents in the projectTry:List agents with their reserved paths
• agent_get

  Get detailed information about a specific agent

  Try:Get details for agent with slug 'feature-x'Try:Show me the coordination brief for agent 'bugfix-42'
• knowledge_list

  List knowledge entries by kind (fundamental or situational)

  Try:Show me all fundamental knowledge for the projectTry:List all situational lessons learned
• knowledge_get

  Retrieve the full body of a specific knowledge entry

  Try:Get the full content of knowledge entry 'auth-guard-order'Try:Show me the lesson about webhook retries
• knowledge_search

  Search knowledge base with natural language queries

  Try:Search for info about rate limit middleware in NestJSTry:Find lessons about handling Stripe webhook retries
• knowledge_create

  Add a new lesson to the knowledge base

  Try:Record a lesson about auth guard ordering in NestJSTry:Save this as knowledge: 'Always validate webhook signatures before processing'
• agent_create

  Create a new agent with reserved paths and implementation details

  Try:Create a new agent to work on the payment featureTry:Dispatch an agent to fix the database migration issue
• agent_update

  Update an agent's status or information

  Try:Mark agent 'feature-x' as activeTry:Update agent 'bugfix-42' with merged commit notes

Comparable tools

Installation

docker compose -f docker-compose.dev.yml up

pnpm install
pnpm dev

{
  "mcpServers": {
    "continuum": {
      "transport": "streamable-http",
      "url": "http://127.0.0.1:7776/mcp"
    }
  }
}

FAQ

What is the difference between fundamental and situational knowledge?

Fundamental knowledge contains project-wide rules loaded on every dispatch, while situational knowledge contains specific insights that are only retrieved when semantically relevant to the current task.

Can I use custom embedding models?

Yes, Continuum supports pluggable embedders including Ollama, Hugging Face TEI, and any OpenAI-compatible service via the EMBEDDER_URL environment variable.

Compare continuum with