Vexa
Meeting bot and transcripts for Google Meet, Teams and Zoom. Live or after, speakers labelled.
Description from the maker (original English)
Multi-account Gmail MCP server — read, search, draft, and label across every inbox.
These three are never merged into one score: opinions, measurements and popularity are different things.
We have not written a plain-language summary for this one yet. Below are the facts we read from the registry.
Pick the AI app you use, then copy the settings.
GMAIL_MCP_CLIENT_SECRET, GMAIL_MCP_DB, GMAIL_MCP_OAUTH_PORTYou need
Which app do you use?
Pick your app above. We remember it, so next time the button says “Install in …”.
Open Claude Desktop → Settings → Developer → Edit Config, and paste this into claude_desktop_config.json
{
"mcpServers": {
"multi-account-gmail": {
"command": "uvx",
"args": [
"multi-account-gmail-mcp"
],
"env": {
"GMAIL_MCP_CLIENT_SECRET": "your-value",
"GMAIL_MCP_DB": "your-value",
"GMAIL_MCP_OAUTH_PORT": "your-value"
}
}
}
}Run this in your terminal
claude mcp add multi-account-gmail --env GMAIL_MCP_CLIENT_SECRET=your-value --env GMAIL_MCP_DB=your-value --env GMAIL_MCP_OAUTH_PORT=your-value -- uvx multi-account-gmail-mcpFill in the empty values before running it.
Paste into .cursor/mcp.json in your project, or ~/.cursor/mcp.json
{
"mcpServers": {
"multi-account-gmail": {
"command": "uvx",
"args": [
"multi-account-gmail-mcp"
],
"env": {
"GMAIL_MCP_CLIENT_SECRET": "your-value",
"GMAIL_MCP_DB": "your-value",
"GMAIL_MCP_OAUTH_PORT": "your-value"
}
}
}
}Your browser asks to open the app, then the app shows the settings and asks you to confirm. Key values are left empty for you to fill in.
Paste into .vscode/mcp.json in your project
{
"servers": {
"multi-account-gmail": {
"type": "stdio",
"command": "uvx",
"args": [
"multi-account-gmail-mcp"
],
"env": {
"GMAIL_MCP_CLIENT_SECRET": "your-value",
"GMAIL_MCP_DB": "your-value",
"GMAIL_MCP_OAUTH_PORT": "your-value"
}
}
}
}Your browser asks to open the app, then the app shows the settings and asks you to confirm. Key values are left empty for you to fill in.
This one runs on your computer, and ChatGPT can only connect to servers on the internet. Pick another app.
Your choice stays on this device in a cookie; no account needed.
GMAIL_MCP_CLIENT_SECRET envGMAIL_MCP_DB envGMAIL_MCP_OAUTH_PORT envThese settings are generated from the maker’s registry entry. If the README says otherwise, follow the README.
Runs on your computer; the tool list appears in your AI app after you install it.
We check information anyone can see; we have not read the code. Checks that do not apply are left out of the count. · Checked Nov 6, 2026
The registry name was verified with the GitHub account cunicopia-dev, which owns the repo
We have not read this repo from GitHub yet
We have not read this repo from GitHub yet
This one runs on your computer; there is no internet endpoint
Runs on your computer and uses local settings instead
Runs on your computer; we can only read tool lists from internet endpoints
Compared with every active server in this category in the registry. Stars count only when the publisher owns the repo.
It passed 1 of 1 applicable basic checks. We have not read the code. Before adding a key or token, see what access the server asks for and use a key with only the permissions you need.
The registry does not list a required key, though some servers ask you to sign in on first use.
It runs on your computer. Your AI app starts it when needed, and data does not pass through the maker’s servers (unless it calls an online service itself).
The GitHub repo last changed –, and the latest version (0.3.0) reached the registry 4 months ago.
Apps that support MCP, such as Claude Desktop, Claude Code, Cursor and VS Code. You also need Python (uv) on your computer.
In the same category: Vexa, emailmd, MCP WhatsApp Cloud
Servers in the same category from other makers, sorted by stars.
Reading a card:★ GitHub starsiThree pillars: reviews · checks · usageiUpdated = latest version published to the registryi
Meeting bot and transcripts for Google Meet, Teams and Zoom. Live or after, speakers labelled.
Render markdown into email-safe HTML, lint drafts for deliverability problems, and preview emails.
MCP server for WhatsApp Cloud API (Meta) — messages, media, and templates via Graph API
Copied from the maker’s repo as written. PopNx MCP has not reviewed or endorsed it. HTML, images and scripts are removed. Links lead off this site.
An [MCP](https://modelcontextprotocol.io) server that reads across _all_ your Gmail accounts from one connection.
Most Gmail integrations — including the native connectors — bind a single account per OAuth grant: connect a second inbox and you disconnect the first. gmail-mcp keeps any number of accounts authorized at once. One Google Cloud client authorizes them all, each lands as a row in a local SQLite file, and every tool takes an account argument that routes to the right mailbox. search_all_accounts sweeps all of them in a single query.
Python 3.12+ · MIT · stdio MCP server + auth CLI · local SQLite token store
It's built to be owned completely: runs in-process over stdio, stores tokens in one SQLite file you can inspect, copy, or delete, talks only to Google and your MCP client, and hardcodes no secrets.
It reads, searches, drafts, and labels. It doesn't send — create_draft leaves a draft for you to send yourself. That's a deliberate default (reasoning in Security model), not a hard stance; if you want autonomous send, it's a small addition or a different server.
Authorize N accounts once via the CLI. Then every tool takes an account, and search_all_accounts hits all of them at once:
search_all_accounts(query="invoice newer_than:30d")
── personal@gmail.com ─────────────────────────────── from: billing@acme.com subject: Invoice #4821 (id 18f...)
── work@company.com ───────────────────────────────── from: ap@vendor.io subject: March invoice (id 19a...)
One query, every inbox, each result tagged with its account and carrying the message id — so the agent can chain read_message(account, id) or create_draft(...) next.
One OAuth client, many inboxes. A single Google Cloud project and one client_secret.json authorize every account. Adding the tenth inbox is the same one-command flow as the first.
Boring storage. Tokens live in one SQLite file under ~/.gmail-mcp/. No daemon, no keyring dependency, no cloud. Back it up by copying it; revoke an account by deleting a row; inspect it with any SQLite tool.
Least privilege. Four granular scopes — gmail.readonly, gmail.compose, gmail.modify, gmail.settings.basic — never the full-mailbox https://mail.google.com/. It can read, draft, label, and manage filters; it never sends mail, and filters it creates can't forward mail off-account.
Headless-friendly. The auth flow assumes the server may have no browser: it prints a consent URL, binds a fixed port, and you SSH-forward the redirect. Works fine on a desktop too.
Every tool except list_accounts and search_all_accounts takes an account (the email address). Unknown accounts return an error listing the authorized ones.
| Tool | Arguments | Returns | ||
|---|---|---|---|---|
list_accounts | — | Authorized accounts + last-used time. Discover valid account values. | ||
search_messages | account, query, max_results=20 | Message summaries (Gmail search syntax) with ids. | ||
read_message | account, message_id, format="full", max_body_chars? | Decoded headers, plaintext body (HTML stripped if needed), attachment metadata. Body capped by default; pass max_body_chars=0 for the full body. | ||
read_thread | account, thread_id, max_body_chars? | Every message in the thread, in order. Each body capped by default; max_body_chars=0 for full. | ||
download_attachments | account, message_id, index? | Save a message's attachments to disk and return absolute paths. Address them by the #N shown in read_message; omit index for all of them. Fixed download root, no destination argument. Dangerous file types and anything on a spam-labeled message are refused. | ||
search_all_accounts | query, max_results_per_account=10 | One search across every account, each result tagged by account. | ||
create_draft | account, body, to?, subject?, cc?, bcc?, html=false, reply_to_message_id?, reply_all=false, from_addr? | A draft (not sent). Returns the draft id. With reply_to_message_id the draft is a reply inside that message's thread: recipient, subject, In-Reply-To, References and the thread id come from it, and to/subject become optional overrides. Without it, to and subject are required. from_addr sets the From header for a verified send-as alias; it defaults to the account address. | ||
list_drafts | account, max_results=20 | Draft ids in the account. | ||
list_labels | account | The account's labels (name + id). | ||
create_label | account, name | Create a user label (nest with /). Idempotent: an existing name (case-insensitive) returns its id and creates nothing. | ||
modify_labels | account, selection (message_id \ | message_ids \ | query), add?, remove? | Add/remove labels on a selection (one id, a list, or everything a query matches), batched 1000/call. General mutator: archive = remove INBOX, mark-read = remove UNREAD, star = add STARRED. |
trash | account, selection (message_id \ | message_ids \ | query) | Move a selection to Trash (recoverable 30 days; not permanent delete). Refuses an empty selection. |
bulk_action | account, action, selection (message_id \ | message_ids \ | query) | Friendly verb layer over modify_labels. action ∈ archive/unarchive/mark_read/mark_unread/star/unstar/spam/unspam/trash/untrash. Batched 1000/call; refuses an empty selection. |
read_messages | account, message_ids \ | query, max_results=25 | Batch-read full content of many messages in one call (vs. N read_message calls). | |
count_messages | query, account?, all_accounts=false | Count matches without fetching content — blast-radius check before a bulk action. all_accounts gives a per-account breakdown + total. | ||
list_filters | account | The account's filters: id, criteria, actions (label ids shown as names). | ||
create_filter | account, one of from_address/to_address/subject/query/has_attachment, plus an action (archive/mark_read/delete/star or add_labels/remove_labels) | A server-side rule applied to incoming mail. Can't forward off-account. | ||
delete_filter | account, filter_id | Remove a filter by id (leaves already-acted-on mail alone). |
flowchart TD
subgraph client[Your machine] Agent[MCP client / agent] CLI[gmail-mcp-auth CLI] Server[gmail-mcp stdio server] Store[("SQLite ~/.gmail-mcp/tokens.db")] Secret["client_secret.json one OAuth client"] end Google[Google OAuth + Gmail API]
CLI -->|"loopback OAuth, once per account"| Google CLI -->|"store refresh token"| Store Secret -.-> CLI Agent -->|"tool call (account=...)"| Server Server -->|"look up + refresh creds"| Store Secret -.-> Server Server -->|"read / draft / label"| Google Server --> Agent
Authorization happens once per account through the CLI (it needs a browser). After that the stdio server reads tokens straight from SQLite, refreshing access tokens on demand and persisting them back. The rest of this section is the "why it works the way it does" detail.
How gmail-mcp authenticates to Gmail, juggles multiple accounts under a single OAuth client, refreshes tokens over time, and authorizes accounts on a headless server. If you just want to get running, jump to Quickstart.
gmail-mcp authenticates using a Google "Desktop app" OAuth client (an *installed application* in OAuth 2.0 terms), driven by the InstalledAppFlow helper from google-auth-oauthlib.
Why an installed-app / desktop client. Installed apps run on a machine the end user controls, so OAuth treats them as public clients: the client_secret in the downloaded client_secret.json is *not* assumed to be confidential. That's the right trust model for a local CLI/desktop tool — there's no server-side component that could keep a secret truly secret, and security rests on the user controlling the redirect (the loopback address) rather than on secret confidentiality. It's the client type Google recommends for command-line and desktop tools.
The loopback redirect flow. After you approve consent in a browser, Google redirects the authorization code to http://localhost: /, where a tiny throwaway HTTP server (started by InstalledAppFlow.run_local_server) catches it. gmail-mcp pins this to a fixed port (default 8765, override with GMAIL_MCP_OAUTH_PORT) and runs with open_browser=False so it works on machines with no browser — see The headless auth path.
Scopes requested. Four granular scopes — never the full-mailbox https://mail.google.com/:
| Scope | What it grants |
|---|---|
gmail.readonly | Read mail and metadata: search messages/threads, read bodies, list labels and drafts. Read-only — cannot modify anything. |
gmail.compose | Create, update, and manage drafts. Used only by create_draft. |
gmail.modify | Add/remove labels on messages and create labels. Used by modify_labels, bulk_action, trash and create_label. |
gmail.settings.basic | List, create, and delete filters. Used by list_filters/create_filter/delete_filter. Does not grant forwarding-address changes (that's gmail.settings.sharing, not requested). |
gmail.send is not requested. Without it the credential simply has no Gmail API path to send mail — the drafts-only behavior is a property of the grant, not just an omitted tool. gmail.settings.sharing is likewise not requested, so no filter can forward mail to another address. The scope list lives in one place: SCOPES in src/gmail_mcp/config.py.
Adding the filter scope to an existing install: widening
SCOPESdoes not
retro-grant already-authorized accounts. Each account must re-run
gmail-mcp-auth addto re-consent to the new scope; until it does, the filter
tools return a
403 insufficient scopeerror.
client_secret.json can authorize any number of accounts.accounts table (~/.gmail-mcp/tokens.db, override with GMAIL_MCP_DB), keyed by email. The row holds the long-lived refresh token, the most recent access-token blob, the granted scopes, and timestamps.Only the beginning is shown; read the rest in the repo