MCP Servers (the outside tools your sessions can call)
An MCP server is an external tool process a session can talk to, and the MCP Servers view is where you manage your own, install official ones from a catalog, and decide which servers each session actually gets. It also tracks how often each one is really used, so unused servers are easy to spot and remove.
On this page
What it is
What it is — An MCP server
An MCP server (Model Context Protocol) is an external tool process Claude Code can talk to — Playwright, Gcloud, a third-party server, anything that speaks the MCP spec. The MCP Servers sidebar view does three things: (1) manages your own custom servers (add / edit / delete), (2) shows per-server usage (count, last-used, tool breakdown) so you can spot dead weight, and (3) is the home of the + Custom manager. Whether a server reaches a given session is decided by the per-project/per-session control layered on top of the global default — see "Controlling which servers a session gets" below.
Where to find it
The MCP Servers entry in the sidebar — the manager and the usage list live behind it, and + Custom starts a new server. A few built-in servers are switched on from Settings, next to the features that use them.
How it behaves
No Omniscio-built MCP server is composed into a session's .mcp.json any more. MemPalace, KMS, Google Workspace, Fathom, Image Studio and web-app verification were each removed from per-session composition; their launch toggles are gone from Settings. Their capabilities are not: the KMS Vault is reached through the control server's /kms/* routes (and the vault skill), MemPalace through its own routes and feature flag, and the rest through their own routes/CLIs — nothing needs a per-session tool process. What still composes is the third-party set below plus your own custom servers.
Controlling which servers a session gets
Each MCP server resolves on/off most-specific-first: session override → project default → global default. Nothing changes for existing setups unless you set an override.
- Global default — the Settings toggles (Zapier, Playwright, Canva, Mobbin) + each custom server's "On by default". These are the default for all sessions. "Off" = installed-but-dormant. Google Drive's MCP is the exception — it never composes. The toggle exists and resolves, but the orchestrator hard-disables the Drive server (
DRIVE_MCP_TIER3_DISABLED = trueinsrc/main/services/mcp/mcp-config-orchestrator.ts:146-150: the read-only Drive MCP needs a Google restricted scope the app does not request, so it would 403), so it is off for every session regardless of the toggle, a project default, or a session override. - Archil is the exception in the other direction — its row is a real switch with no MCP server behind it. Switching Archil on composes nothing; it rewrites how the session starts so its file, search and command tools run in Archil cloud microVMs (archil-workspace.md).
- Per project — Edit a project → MCP tools: flip each tool on or off for that project's sessions.
- Per session — override for just that session, from two entry points: the session's ⋯ menu → MCP servers any time, OR — on a brand-new session — the "MCP servers" button in the launch-config block, right next to the provider/model/thinking pickers. Both open the same on/off switch editor (titled "MCP tools for this session") and write the same per-session override; it applies on the session's next start (MCP servers load when the CLI launches), so choosing at launch takes effect on the first message.
- Add your Fathom key — while no Fathom key is saved, the Fathom row in that switch editor shows an Add your Fathom key link. It opens Settings → My API Keys at the Fathom key. The link disappears once a key is saved. Fathom's key is the same one saved on the Mission Control Meeting notetakers card, so you only ever have one. Fireflies has no link here: its server is a separate install from the catalog with its own browser sign-in.
Scope covered: the composed built-ins (Zapier, Playwright, Canva, Mobbin) + all custom servers. Google Drive is resolvable but hard-disabled in the orchestrator, so it never composes (see the global-default bullet above).
Adding a custom server
- MCP Servers sidebar → + Custom opens the manager.
- Add server — name, key (the
.mcp.jsonentry name), type (localstdiocommand, or remotehttp/sseURL), args, env vars / headers (secrets), and "On by default". - Secrets are encrypted at rest and never shown back (you re-enter to change them). A custom server is off by default until you switch it on per project/session.
- Edit / delete from the same manager. Delete is confirm-gated.
Install an official server (catalog)
The Add MCP server dialog opens with a searchable catalog of 80+ official remote servers — Nous Research's Hermes Agent set plus official servers we independently verified beyond it (GitHub, HubSpot, Firecrawl, Perplexity, PagerDuty, Box, Zoom, and more; Linear, Notion, Figma, GitLab, Stripe, Datadog, Grafana, Sentry, Supabase, Vercel, Netlify, Cloudflare, Atlassian, Asana, Intercom, HuggingFace) — so you never have to hunt down a URL. It works like a plugin marketplace: nothing is installed until you choose it. The full, verified list lives in mcp-install-catalog.ts.
- Browse or search the list, then click Install on a server to add it — the row flips to Installed once it's in your list, and you can install several without closing the dialog.
- Each row shows the service's own logo, so you can spot the one you want at a glance. The artwork is bundled with the app (public-domain Simple Icons path data, generated into mcp-brand-icons.generated.ts by
npm run mcp-icons:reindex) — the dialog never calls out to a logo service, so browsing the catalog leaks nothing about which servers you're looking at and works offline. A service with no bundled mark shows a neutral glyph instead: a missing logo is fine, a wrong one is not. - Nothing is added by default. The catalog is just a list of what's available; installing is an explicit, per-item choice.
- Same add path. Install just pre-fills the name, remote URL, and transport, then runs the exact
addManagedMcp()a manual server uses — so a catalog server behaves identically (full per-project/per-session control, delete from the same manager). - Sign in from the server's page. Omniscio's own sessions cannot run a connector's browser sign-in, so a remote server (http/sse) shows a Signed in or Needs sign-in label and a Sign in button on its Configuration card. Pressing it runs Claude Code's own
claude mcp loginin a private folder (<userData>/mcp-signin/), opens the sign-in page in your browser, and shows "Finish signing in in your browser" until you approve (it gives up after 10 minutes). The server is registered only in that private folder, so no other session gains it; once signed in, every Omniscio session that has the server enabled uses the sign-in with no prompt. Sign out runsclaude mcp logout. Only one sign-in can be open at a time. - Installing only registers the server. A catalog entry does nothing until you install it, and a remote server still needs its sign-in (above). Upwork is in the catalog for Writer Studio's client conversations. GitLab's endpoint is Beta and needs a GitLab admin to enable it for your instance first (ClickUp is public beta); both caveats are shown on the row.
- The catalog is a small curated data file, mcp-install-catalog.ts — every URL is verified against the vendor's own docs (a wrong URL is worse than no entry). Browse UI:
McpCatalogBrowserin mcp-servers/.
From the CLI — add any catalog server over the control server by POSTing its entry to the standard add route POST /managed-mcp (full-trust amc-cli token — the add route is secret-capable). Example (Linear):
node "$HOME/.claude/skill-content/omniscio-control/amc-api.mjs" POST managed-mcp \
'{"key":"linear","displayName":"Linear","transport":"http","url":"https://mcp.linear.app/mcp"}'
Look up any service's key / transport / url in mcp-install-catalog.ts.
Sign in or out of a remote managed server from the CLI (full-trust token for the two writes):
POST /managed-mcp/:id/sign-in- runs the browser sign-in and answers when the person finishes (up to 10 minutes). A failure answers with a plainerrorand areason(canceled,timed-out,refused,claude-missing,busy); a server that runs a command on this computer (stdio) is refused with400.GET /managed-mcp/:id/sign-in-status-data.statusisconnected,needs-sign-inorunknown.POST /managed-mcp/:id/sign-out- clears the sign-in.
A per-project switch for a server in a built-in area such as Writer Studio is written against that area's own project path (for Writer Studio, __writer__) through POST /managed-mcp/:id/project-override, and it applies to that area's sessions.
Tool controls (Lab)
Behind Settings → Lab → MCP tool controls, a server you select in the MCP Servers view shows a Tool controls panel with three things. They apply to your own custom servers and to managed ones, and they are global for the server (not per project).
- Switch off a tool. Each tool the server has been seen using has a Switch off control, and Add a tool name switches off one that is not in the list yet (use the tool's own name, letters, numbers, dots, dashes and underscores, up to 128 characters, up to 200 per server). A tool you switch off is not offered to the agent at all. This reaches Claude and Codex sessions; other engines ignore it. The panel says which engines each control reaches: always ask and the call limit are Claude only. It takes effect on the session's next start, like any change to a session's servers.
- Trust: Standard or Always ask me. Standard is today's behaviour. Always ask me puts a question in front of you for every call to that server, at every permission level, even where your own settings or Full trust would let the call through, and no earlier "always allow" answers it. This is for Claude sessions. A cloud session cannot ask you, so a server set to always ask is left out of a cloud session's tools, and the session says so in its transcript.
- Calls per session. A cap (1 to 10,000, or no limit) on how many times one session may call the server. When the cap is reached, further calls are refused with a plain message naming the server and the limit. This is for Claude sessions. The count belongs to the session, is kept in memory and is forgotten when the session ends. The cap holds where Omniscio is asked to approve a call, which means a server set to always ask or a permission level that asks; a call your own settings let through without asking cannot be stopped.
The panel counts calls, never dollars: Omniscio measures cost per turn, not per tool.
With the Lab switch off, nothing is blocked, asked or counted.
When an MCP server asks you a question
Some MCP servers stop mid-task to ask the person something, for example to fill in a small form or to open a web page to sign in. In a Claude session this now appears as a card in the session panel, titled Question from {server}, next to the permission card. It is not behind the Lab; before this, such a request was silently cancelled.
- Form questions show the fields the server asked for: text, number, whole number, yes/no and a choice from a list, with required fields marked. Submit sends your answer and Decline refuses. If the server asks for something the card cannot show, the card says so and offers only Decline.
- Web page questions show the page's host and an Open button that opens it in your browser (only http and https addresses, only when you press the button). Then press Done.
- Nobody answers. The card cancels itself at the approval timeout you set under Safety, or after ten minutes when none is set. A card still open when its session ends is cancelled.
- Sessions that cannot ask you (scheduled runs, chat gateway sessions and cloud sessions) decline at once, so they never wait for an answer that cannot come.
- On the phone the card shows the question and tells you to answer it on your computer; a phone cannot submit, decline or open the page.
- Not covered. Codex and other engines do not pass these questions to Omniscio, and no engine passes the "sampling" requests a server can make, so nothing is shown for them.
Usage tracking (unchanged)
- Sort — Most / Least / Never used / Recently used / Alphabetic. "Never used" is the "what should I uninstall?" lens.
- Per-tool breakdown — click a server to see its per-tool invocation table.
- Reset usage — right-click a row (
UsageRowContextMenu); footer Clear usage wipes all, both confirm-gated. - Scan — forces an immediate re-scan of
~/.claude/projects/**/*.jsonl.
Related
Related — use-skills.md
- use-skills.md — sibling Skills sidebar (same usage model)
- zapier-integration.md — the composed built-in MCP servers that remain
- kms-agent-tools.md — the Vault's agent surface, now served entirely by the control server's
/kms/*routes (no per-session MCP) - canva-integration.md — Canva connected as an MCP server: the one-toggle setup and the roughly thirty design tools an agent gains
- archil-workspace.md — the other direction: a session's own file, search and command tools run in Archil cloud microVMs instead of on your computer
Last verified 2026-11-16