Stash / Blog · June 2026 · 7 min read
Claude Projects vs MCP Connectors: What's the Difference?
Claude has two distinct ways to carry context beyond a single conversation: Projects and MCP connectors. Both extend what Claude can remember. They work differently, suit different jobs, and — used together — they cover most of what power users actually need.
This piece explains each one plainly, then shows when to reach for which.
Claude Projects
Projects are a Claude.ai feature (available on Pro and higher plans). A Project is a container you create in the sidebar. It has three things:
- A persistent system prompt — instructions Claude follows in every conversation inside the project, automatically, without you typing them again.
- Uploaded files — documents, PDFs, and text files Claude can refer to throughout the project's lifetime.
- Shared conversation history — conversations inside a project are grouped. Claude surfaces relevant context from past chats when you start a new one in the same project.
The key word is manual. You write the system prompt. You upload the files. You organise the conversations. Projects are great at holding a stable brief: "You are my copywriting assistant. Our brand voice is X. Here are our guidelines."
What Projects are good for
- Role-playing a consistent persona across many sessions
- Uploading a document (spec, brief, style guide) that Claude should reference
- Keeping a work-in-progress project (a book chapter, a code module) organised in one place
- Team access: multiple Claude users sharing the same project context
What Projects don't do
- Live data access — your uploaded files are static snapshots. Claude cannot query an external database, check a live list, or read something that changed after upload.
- Programmatic updates — you cannot append to a Project document from a script. If your contacts list grows by 20 names this week, you upload a new version manually.
- Token-light retrieval — the whole file goes into context on every read. A 500-row CSV loaded at the start of each conversation costs the same whether Claude needed all 500 rows or just one.
MCP connectors
MCP (Model Context Protocol) is an open protocol that lets Claude connect to external services at conversation time. An MCP connector is a URL you add to Claude once, in Settings → Integrations. After that, Claude can call tools the connector exposes — read records, search data, write new entries — as part of any normal conversation.
The key word is live. The data lives outside Claude, in whatever service the connector points to. Claude fetches exactly what it needs, when it needs it.
What MCP connectors are good for
- Persistent, queryable records — contacts, tasks, notes, decisions, anything Claude should be able to search on demand
- Data that changes — a connector reflects the current state of your data, not a snapshot from last week
- Token-efficient retrieval — instead of loading a whole dataset, Claude calls
search(collection="contacts", q="Sarah Chen") and gets one record
- Programmatic writes — your scripts can add records to the connector without you touching Claude at all
- Cross-conversation consistency — every conversation (in any project, or none) can call the same connector
What MCP connectors don't do on their own
- Persistent instructions — a connector gives Claude tools, not a persona. If you want Claude to always behave as your personal assistant, that still belongs in a Project system prompt (or custom instructions).
- File storage — most record-store connectors (including Stash) store structured text records, not arbitrary files. PDFs and docs still belong in a Project.
Side by side
| Feature |
Claude Projects |
MCP connector |
| Persistent instructions for Claude |
✓ (system prompt) |
✗ |
| Uploaded static documents |
✓ |
✗ |
| Live external data |
✗ |
✓ |
| Programmatic writes |
✗ |
✓ (connector-dependent) |
| Token-light search/retrieval |
✗ (whole file in context) |
✓ (fetch only what's needed) |
| Cross-project availability |
✗ (scoped to project) |
✓ (available everywhere) |
| Team sharing |
✓ (Pro+) |
Depends on connector |
The smart play: use both
Projects and MCP connectors solve different halves of the same problem. Projects handle who Claude is for a task. MCP handles what Claude knows at runtime.
A concrete setup that works well:
- Create a Claude Project called Work assistant. System prompt: your role, your priorities, how you like replies formatted, the names of your colleagues.
- Add your MCP connector (e.g. Stash). Now Claude can search your contact list, pull up meeting notes, or check your task list — without you pasting anything.
The Project provides the stable brief. The MCP connector provides the live data. Neither one alone is as useful as both together.
Tip: Put things that don't change in your Project — your role, your writing style, your team names. Put things that grow or change in your MCP connector — contacts, tasks, decisions, notes. The rule: if you'd have to re-upload it to keep it current, it belongs in MCP.
Which MCP connector for persistent records?
There are now dozens of MCP connectors available. For persistent record storage specifically — the "Claude knows my data" use case — you want a connector that:
- Runs remotely (no local server to manage)
- Has full-text search (so Claude can fetch one record, not 500)
- Requires no code to set up
- Works immediately with your existing Claude account
That's what Stash is built for. You add the connector URL, sign in with Google, and you have a searchable record store that Claude can read and write to from any conversation. Free to start.
Add Stash to Claude in 30 seconds
Paste the connector URL into Claude → Settings → Integrations. Sign in with Google. Done.
Get your connector URL →
Frequently asked questions
Can I use MCP inside a Claude Project?
Yes. MCP connectors are account-level — they're available in every conversation, including those inside Projects. Adding Stash to your Claude account means you can call it from any Project, or from no Project at all.
Do I need a paid Claude plan for MCP?
MCP connector support is available on Claude.ai across plans. Check the current Anthropic docs for your specific plan's limits — plans and features change over time.
What if I already have a lot of data in a Project?
Projects and MCP can coexist. You might keep your static reference docs (a brand guide, a spec) in a Project, and migrate your growing lists (contacts, tasks, decisions) to an MCP connector like Stash. You don't have to choose one or the other.
Is MCP only for developers?
Remote MCP connectors (like Stash) require no code to set up. Local MCP servers do require some technical setup. If you're not a developer, stick to remote connectors — you paste a URL, that's it.
Can multiple Claude Projects use the same MCP connector?
Yes. Because MCP connectors are registered at the account level, not the project level, they're available in every Project. One Stash account serves all your Claude Projects.