Get This Tool
Reference MCP
Pricing
- Model
- Free
Summary
Claude Code forgets what you decided in Codex. Codex can't see your Cursor history. Every agent starts from scratch, and you spend the first ten minutes of every session re-explaining context that already exists on your machine.
Reference is a local MCP server that reads session transcripts and memory files — CLAUDE.md, AGENTS.md, and similar — from multiple AI coding tools, then exposes that history to whichever agent is asking. Register it once in each tool's MCP config and any agent can query what any other agent did before. The architecture is passive: Reference does not plan, execute, or chain tasks — it answers lookups. That scope is the point, and also the ceiling. Teams with more than a handful of tools, or who need structured, queryable memory rather than transcript search, will find the file-scanning approach starts to feel fragile as session volume grows.
Bottom line: Pick this if you are context-switching between Claude Code and Codex daily and tired of re-explaining last week's architecture decisions — skip it if you need memory that survives across machines or integrates with a team.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Community Benchmarks Community
Sign in to submit a benchmarkNo community benchmarks yet. Be the first to share a real-world data point.
Pros
Sign in to edit- Reads session transcripts from Claude Code, Codex CLI, and Cursor in one place, so you stop re-explaining decisions that were already made in a different tool last week.
- Fully local and offline, which means code and conversation history never leave your machine — critical for projects where sending context to a third-party sync service is off the table.
- MIT-licensed with no paid tiers and no account required, so there is no vendor relationship to manage and no access cliff if a pricing tier changes.
- Single MCP registration pattern works across supported tools, so you configure it once per tool rather than wiring a custom integration for each pair.
Cons
Sign in to edit- Search runs over raw transcript files and markdown — not a structured index. Once session volume is high enough that the answer you need is buried in dozens of transcripts, recall becomes a best-effort grep rather than a reliable query, and teams at that point look at purpose-built vector stores like a local Chroma or Weaviate instance instead.
- Memory is local to one machine. A team of two engineers running the same agents on different laptops gets zero shared context — Reference has no sync layer. Teams who need shared agent memory across contributors abandon this and wire a shared database backend, at which point Reference's architecture no longer fits.
- There is no API and no programmatic query surface outside the MCP protocol. Any workflow that needs to pull agent history into a dashboard, a CI pipeline, or a custom tool has no supported path — the docs describe no extension point for that use case.
Community Reviews
Sign in to write a reviewNo reviews yet. Be the first to share your experience.
About
- Platforms
- Python / local
- API Available
- No
- Self-Hosted
- Yes
- Last Updated
- 2026-07-01T02:35:18.264Z
Best For
Who it's for
- Users running multiple AI coding agents
- Lightweight local memory servers
What it does well
- Cross-tool session recall for Claude Code and Codex CLI
- Agent memory aggregation from multiple IDEs and CLIs
- Local offline search of past AI interactions
Integrations
Discussion Community
Sign in to commentNo discussion yet. Sign in to start the conversation.
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeCommunity Notes & Tips Community
Sign in to contributeBe the first to contribute. General notes, observations, gotchas, and tips from people who use this tool day-to-day.
Frequently Asked Questions
- Is Reference MCP free?
- Yes — Reference MCP is fully free to use. There is no paid tier.
- Is Reference MCP open source?
- Yes. Reference MCP is open source.
- Can I self-host Reference MCP?
- Yes. Reference MCP supports self-hosting on your own infrastructure.
- What platforms does Reference MCP support?
- Reference MCP is available on: Python / local.
Hours Saved & ROI Stories Community
Sign in to contributeBe the first to contribute. Concrete time/cost savings, with context. e.g. "Cut my code review backlog from 4h to 45m per week."
Curated lists that include this category
When you run Claude Code for one project and Codex CLI for another, neither tool knows what the other decided. Reference solves that by acting as a shared memory layer: an MCP server you register in each tool that reads session transcripts and memory files from your local filesystem and makes them searchable by any agent that queries it. The workflow is install-once, configure-once — add it to each tool’s MCP config via uvx or a git clone, and agents can call it whenever they need to recall what a sibling tool resolved in a past session.
The differentiating move here is cross-tool recall without a cloud backend. All reads happen locally, offline, against files already on your machine. There is no sync service, no account, no data leaving your system. For developers who have hard requirements around code privacy or work in air-gapped environments, that architecture is not a compromise — it is the feature.
Where Reference fits cleanly: solo developers running two or three AI coding agents who want shared context without standing up infrastructure. Where it breaks: the search is over raw transcripts and markdown files, not a structured knowledge graph. As session volume grows, recall quality depends on how consistently your tools write session files and how well the agent interprets unstructured history. Teams who need memory shared across machines, or who want to query decisions with precision — ‘what did we decide about the auth layer in sprint three’ — will find transcript scanning insufficient and reach for a purpose-built knowledge store.
Reference is MIT-licensed and self-hosted by definition — there is no hosted version. Installation uses standard Python tooling (uvx or pip from the repo), and the example config file covers the MCP registration pattern for supported tools.