ResumeContext

Product Introduction

1. Definition: ResumeContext is a specialized MCP (Model Context Protocol) server designed for AI-powered coding assistants. It functions as a persistent, shared memory archive that automatically captures, syncs, and serves the session history from tools like Claude Code, Cursor, and Codex across an entire development team.
2. Core Value Proposition: It exists to solve the critical problem of context fragmentation in AI-assisted software development. Its primary value is enabling persistent project memory that survives individual agent sessions and user accounts, preventing knowledge loss and work duplication. This directly addresses the high switching cost between AI coding agents and breaks down team knowledge silos trapped on individual laptops.

Main Features

1. Universal Session Archiving: The system operates via a background daemon (CLI tool) that reads the local session history files written natively by supported coding agents. It requires no plugins or changes to the agent interface itself. The daemon syncs this data—including prompts, code, agent replies, and tool call outputs—to a centralized, project-scoped archive approximately every 20 seconds. This creates a searchable transcript of all development reasoning.
2. MCP-Powered Context Retrieval: The shared archive is exposed to coding agents exclusively through the Model Context Protocol (MCP). When an agent is connected to the ResumeContext MCP server for a project, it gains tools to query the team's historical sessions. This allows the AI to answer questions like "did anyone fix this bug?" by retrieving and citing specific lines from past sessions, complete with attribution to the original teammate.
3. Agent-Agnostic Project Memory: The archive is fundamentally decoupled from any specific AI agent or user account. It is tied to the project directory (via resumecontext init). This architecture allows a developer to hit a usage limit in one agent (e.g., Claude Code) and seamlessly continue work in another (e.g., Codex or Cursor), with the new agent instantly aware of all prior context through the same MCP connection.
4. Team Intelligence & Management Tools: Beyond simple search, the MCP server provides analytical tools. A team lead can query the archive to get summarized status updates, identify blockers (e.g., "Sam has hit the same failing test four times"), or spot work duplication, all sourced directly from the team's actual session data without requiring manual status reports.

Problems Solved

1. Pain Point: Loss of Institutional Knowledge in AI Sessions. Critical debugging steps, design decisions, and dead-end explorations conducted within an AI coding session are typically ephemeral, locked in a local transcript on one developer's machine.
2. Pain Point: Agent Lock-in and Context Loss on Switching. Hitting a usage limit or choosing to switch AI agents forces developers to start context from scratch, re-explaining the entire task and losing all in-session reasoning.
3. Target Audience: Software development teams actively using multiple AI coding assistants (Claude Code, Cursor, Codex, etc.). Engineering managers and tech leads who need visibility into progress and blockers without interrupting flow. Individual developers who use different agents for different tasks and suffer from context fragmentation.
4. Use Cases: Onboarding new team members, whose AI agent can immediately query the project's history to understand past decisions. Cross-timezone handoffs, where work done by one developer's AI is immediately accessible to another's. Debugging recurring issues, by searching the archive for past investigations and fixes. Continuing complex tasks after an agent's context window is full or a usage limit is reached.

Unique Advantages

1. Differentiation: Unlike screen recording or manual note-taking, ResumeContext automates knowledge capture at the source (the agent session) without developer overhead. Compared to generic team wikis, it captures granular, cited reasoning automatically. Versus single-agent memory features, it is vendor-agnostic, creating a unified memory layer across competing AI coding tools.
2. Key Innovation: Its core innovation is leveraging the Model Context Protocol (MCP) as a standardized conduit for both writing and reading project memory. This turns the MCP from a simple context-injection tool into a bidirectional memory bus for the entire development lifecycle. The architecture of a project-scoped, agent-agnostic archive is a novel approach to persistent context in the AI development toolchain.

Frequently Asked Questions (FAQ)

1. How does ResumeContext work with different AI coding agents like Cursor and Claude Code? ResumeContext uses a lightweight CLI daemon that reads the native session history files each agent writes to your local disk. It requires no plugin within the agent itself. You configure which agents to archive per project, and the daemon syncs their transcripts. The agent then accesses this shared memory through a standard MCP connection.
2. Is my source code stored in the ResumeContext archive? Yes, but in a specific context. The archive stores the full conversation transcript, which includes any source code that was read, generated, or discussed by the AI agent during a session. This data is scoped strictly to the project directory it was recorded from and is only accessible to members invited to that specific project.
3. What happens if a team member leaves the project? The project admin can revoke the member's access immediately. The historical sessions they recorded remain in the archive and stay attributed to them. The archive is not rewritten, preserving the project's complete historical record for the remaining team members.
4. Can ResumeContext see or access my private AI sessions from other projects? No. Access is controlled per project, defined by the directory where resumecontext init was run. Sessions from other directories or projects are not synced unless you explicitly initialize and configure ResumeContext there. The MCP connection for each project is isolated.
5. What is the latency between a local AI session and it being available in the shared archive? The daemon syncs session data approximately every 20 seconds. The primary gap is the current, active session you are in, as it has not yet been synced. Therefore, the archive is best for querying past work, not the immediate last few minutes of your current task.