ccontext

ccontext MCP Server is a local-first execution context for AI agents. It provides a shared, durable memory and structured collaboration space that lets agents resume work across sessions, coordinate in real time, and keep context clean through built‑in diagnostics and lifecycle rules.

How to use

You use ccontext by running the MCP server through an MCP client and then interacting with its functions from your agent code or workflow. Start by loading the current context, then read or update the Vision, Sketch, Milestones, Tasks, Notes, and Presence. Use the one-call commit_updates flow to batch presence, task progress, and notes, so your agents stay in sync with minimal friction. Always check diagnostics to understand context debt and top issues, and rely on the Presence data to see who is doing what.

Begin a run by loading the context so you can orient yourself with version, current milestones and active tasks. If foundational information is missing, set the Vision and Sketch first, then ensure there is an active Milestone. Create Tasks for concrete deliverables and define steps with acceptance criteria. Use batch updates to update presence, advance task steps, and add notes in a single operation to keep the context coherent across agents and sessions.

How to install

Prerequisites: You need a modern Python or Node-based environment depending on how you run the MCP server with your MCP client.

Install using Claude for recommended workflows if you have Claude installed:

# Using uvx (recommended)
claude mcp add ccontext -- uvx ccontext-mcp

# Or using pipx
claude mcp add ccontext -- pipx run ccontext-mcp

Configuration snippets for running with Claude Desktop and manual MCP clients

Claude Desktop users can configure the MCP server in your claude_desktop_config.json as shown below.

{
  "mcpServers": {
    "ccontext": {
      "command": "uvx",
      "args": ["ccontext-mcp"],
      "env": { "CCONTEXT_ROOT": "/path/to/your/project" }
    }
  }
}

Manual / standalone run

If you run the MCP server manually, install the package and run the server using the provided command, adjusting the root path to your project.

# Example manual run command (adjust as needed)
CCONTEXT_ROOT=/path/to/your/project ccontext-mcp

Environment variables

Root directory for the local MCP project is controlled by CCONTEXT_ROOT. Set it to the path where your context/ is stored.

Tools and operations you can perform

The MCP server exposes a set of functions to manage the context across sessions. You can load the current context, update the Vision, Sketch, Milestones, and Tasks, manage Notes/References, and track Presence. Use batch updates to combine presence changes, task progress, and notes in a single operation.

Version tracking and diagnostics

The system exposes a version identifier and a diagnostics object. You can detect changes to the context by comparing versions between reads. Diagnostics include a debt score and top issues to help you keep the context healthy and up to date.

Notes on hygiene and lifecycle

Notes and references decay over time with a TTL-based lifecycle and can auto-archive when stale. Presence data is normalized for readability, and the system highlights any context debt to guide cleanup.

Versioning and change detection

Use the version value from get_context() to detect changes without polling every field. If the version changes, refresh your local view to keep in sync with other agents.

Available tools

get_context

Load the current execution context, including version, time, diagnostics, and the full context.

commit_updates

Batch multiple updates in one call, such as presence status, task progress, and notes/refs.

update_vision

Set or update the north star vision for the autonomous agent.

update_sketch

Update the static blueprint detailing architecture, strategy, and constraints.

create_milestone

Create a new milestone and optionally mark it as active.

update_milestone

Modify details of an existing milestone.

complete_milestone

Mark a milestone as completed.

remove_milestone

Remove a milestone from the context.

list_tasks

List current tasks and their statuses.

create_task

Create a new task with steps and an associated milestone.

update_task

Update task details or steps.

delete_task

Delete a task.

add_note

Add a note with an optional TTL and lifecycle.

update_note

Update an existing note.

remove_note

Remove a note.

add_reference

Add a reference to a file or URL with TTL handling.

update_reference

Update an existing reference.

remove_reference

Remove a reference.

get_presence

Query current agent presence data.

update_my_status

Update your current status in a concise line.

clear_status

Clear your current status.