AI assistant integration (MCP)

What MCP is, in one paragraph

Model Context Protocol (MCP) is an open standard for connecting AI assistants to external tools and data. An assistant connects to Overshow and is offered a set of tools it can call to retrieve your local context. Think of it as a well-defined bridge between a chat interface and a data source, replacing ad-hoc copy-paste with a consistent, auditable contract.

For Overshow, this means an approved assistant can search bounded snippets from your local screen text, meeting transcripts, and indexed documents.

Compared with pasting large raw exports into a chat, tools return bounded, filterable snippets by time, source type, and, for screen results, app.

What MCP lets you do

You want to	The assistant can
Recall what you were working on last Tuesday	Search eligible local text with time and app filters
Find where a topic came up in a meeting	Search confirmed transcript snippets inside eligible meetings
Pull context into an IDE session	Fetch bounded matching snippets and opaque source references

The shared thread: you stay in the assistant you already use, and Overshow becomes one of the grounded sources it can cite.

What the tools cover

The shipping integration exposes exactly three read-only tools. Assistants cannot execute arbitrary code against your data:

• search_memory searches eligible screen text, confirmed meeting transcripts, and indexed document text.
• get_memory_source reads a small source window for an unexpired opaque source id, when you grant source-detail access.
• memory_status reports coarse availability, pairing, scope, and index state without paths, tokens, queries, or content.

It does not expose profiles, graph data, questions, summaries, mail, clipboard history, settings, or write operations.

Privacy model

The integration is designed around a simple rule: where your query and results end up depends on which assistant you connect.

Assistant type	What leaves your machine
Local LLM (for example, Jan.ai or LM Studio)	Nothing. Queries, tool calls, and responses stay on device.
Cloud assistant (Claude, Cursor, others)	Query text and any tool results returned to the assistant are sent to that provider's servers.

Because tool results can include verbatim OCR text, transcripts, and UI snapshots, sending them to a cloud assistant effectively ships that content to a third party. Overshow therefore treats cloud assistants as an explicit opt-in:

• A local client (for example a local LLM connected through the custom flow) keeps the entire exchange on your machine.
• Connecting Claude, Cursor, or similar cloud clients requires an explicit opt-in.
• That choice is auditable in the desktop app so you can see what was enabled and when.

Separately, the server runs on localhost only. It is not exposed to the network. An assistant has to be running on the same machine (or explicitly tunnelled by you) to call it.

How it fits into your machine

When Memory MCP is enabled, an assistant connects to the packaged local helper on demand and reads only what each tool call returns. The desktop app must be running because the helper never opens the encrypted database itself.

Setup

Connect a client from Settings → AI clients. Overshow gives you a helper configuration to add to your assistant; there is no token to copy. The client appears for approval after it first connects.

Step	What you will do
1. Install Overshow	Install the desktop app and complete onboarding as normal.
2. Enable Memory MCP	Turn it on in Settings. Free accounts cannot enable it.
3. Configure and restart	Copy the generated client configuration and restart the assistant.
4. Approve the pairing	Approve the pending client, choose its history limit, and optionally grant source detail.
5. Ask your first question	The assistant can now search eligible memory and cite returned sources.

Every setup requires explicit Overshow consent and per-client approval. A cloud assistant also sends the query and returned snippets to its provider.

What to expect in practice

A few realistic scenarios with the shipping Memory MCP (search_memory, get_memory_source, memory_status), once a client is connected:

Scenario	What happens
You ask Claude about yesterday's standup	The assistant calls search_memory for that meeting and replies from the returned snippets.
Cursor needs the cited passage	After a search hit, the assistant calls get_memory_source (if you granted source detail) for a redacted window.
A local LLM helps you write a follow-up	The assistant searches your memory for recent mentions of the person or topic and drafts from those snippets.
You check whether Memory MCP is ready	The assistant calls memory_status to confirm consent, entitlement, and connection state.

In each case the assistant sees only what the tool returns for that specific call, not the entire database.

Settings lets you change an approved client's history limit later. The effective window is the lower of that client limit and your account limit. Privacy settings can additionally redact email addresses, phone numbers and IBANs from client responses; secret, credential and payment-card patterns are always scrubbed. You can also hide an individual person, organisation or project from connected AI tools from that profile's detail page.

Trade-offs worth knowing now

• Tool results are bounded. Each call returns a limited window (for example, a page of search hits or a capped number of events). Assistants may need to call more than once for broad questions, which costs latency and, on cloud assistants, tokens.
• Grounding depends on retrieval. The assistant's answer is only as good as the captures surfaced by the tools it chose. Vague prompts tend to trigger broad searches with mediocre results.
• Cloud assistants can be verbose. Some clients will read large tool outputs into their context repeatedly. If that matters for cost or privacy, prefer a local LLM for day-to-day work and reserve cloud assistants for tasks that clearly need them.
• MCP is still maturing. Clients differ in how they display tool calls, cite sources, and handle errors. Overshow follows the protocol closely, but the experience also depends on the assistant you pair with it.

What it does not do

• It does not give assistants write access to your Overshow data.
• It does not stream captures in real time; it answers on-demand queries.
• It does not replace the desktop Ask feature. Ask stays the primary grounded-answer surface inside Overshow itself.
• It does not bypass capture controls. Paused or excluded content is not indexed, so MCP cannot retrieve it.

Client support

The Memory MCP works with MCP-capable clients on any Pro or Enterprise account, including the trial. Settings includes named setup presets for Claude Desktop, Claude Code, Cursor, OpenAI CLI, and Perplexity Mac, plus a custom flow for any other MCP-capable client, including local LLMs; the contact page is the best way to tell us which presets to add next.

See also

• Ask: the grounded-answer surface inside Overshow that uses the same local retrieval signals.
• Search: the filters and modes the integration's search tools map onto.
• Privacy: on-device processing, capture controls, and encryption, which together define what the integration can ever see.
• Security: how cloud assistant integrations are treated as an explicit opt-in.