Skip to content
AlbertCode SWE

MCP servers

Model Context Protocol servers give Albert outside tools: Git, a browser, a database, documentation, your company's own services. AlbertCode connects to servers that run on your computer (a local command) and servers on the web (a URL), including ones that need you to sign in.

The quickest way: connectors

Connectors is a gallery of 88 ready-made servers. Each one is added with a click and, at most, the answers it needs: a sign-in for services on the web, a folder, or a database address. Every connector in the gallery has been run for real: started or reached the way AlbertCode does it, with its tools listed and one of them called (each database connector against a real database of its kind). A connector that signs in was taken up to the service's own sign-in page: AlbertCode registered itself with the service, and the rest is your own account.

Code and repositories

Connector (id) What it gives Albert Asks for
Git (git) Read and search a local Git repository: status, diffs, log, branches The repository folder
Filesystem (filesystem) Read, write and search files, only inside the folders you name The folders
GitLab (gitlab) Issues, merge requests, pipelines and code search in your GitLab projects Sign in
Serena (serena) Find symbols, references and definitions through language servers, and edit code by symbol The project folder

Projects and work

Connector (id) What it gives Albert Asks for
Linear (linear) Find, create and update Linear issues, projects and comments Sign in
Notion (notion) Search, read and write Notion pages and databases Sign in
Jira and Confluence (atlassian) Search and update Jira issues and Confluence pages (Atlassian Rovo) Sign in
Asana (asana) Tasks, projects, goals and comments in Asana Sign in
monday.com (monday) Boards, items, updates and workspaces on monday.com Sign in
ClickUp (clickup) Tasks, lists, docs and time tracking in ClickUp Sign in
Todoist (todoist) Tasks, projects, sections and comments in Todoist Sign in
Airtable (airtable) Bases, tables and records in Airtable Sign in
Zapier (zapier) Run actions in thousands of apps through your Zapier account Sign in
Fireflies (fireflies) Search meeting transcripts, summaries and action items from Fireflies Sign in
Connector (id) What it gives Albert Asks for
AWS Documentation (aws-docs) Search and read AWS documentation pages
AWS Knowledge (aws-knowledge) AWS documentation, API references, What's New posts and regional availability, with nothing to install
Microsoft Learn (microsoft-learn) Search and read Microsoft and Azure documentation and code samples
Context7 (context7) Up-to-date documentation and code examples for thousands of libraries
DeepWiki (deepwiki) Ask questions about any public GitHub repository and read its generated wiki
GitMCP (gitmcp) Documentation and code of any public GitHub project, read from the repository itself
Cloudflare Docs (cloudflare-docs) Search Cloudflare's developer documentation
Exa (exa) Web search, page contents and code search built for agents
Fetch (fetch) Fetch a web page and hand it to the agent as Markdown
Hugging Face (huggingface) Search models, datasets, Spaces and papers on the Hugging Face Hub. Sign in for your own Sign in, for your own
Wolfram (wolfram) Wolfram Alpha answers and Wolfram Language computation: maths, science, data and units

Browser and testing

Connector (id) What it gives Albert Asks for
Playwright (playwright) Drive a real browser: open pages, click, type, read what is on screen, take screenshots Needs Chrome
Chrome DevTools (chrome-devtools) Inspect a live Chrome: console, network, performance traces and the DOM Needs Chrome
Cloudflare Browser Rendering (cloudflare-browser) Fetch pages, take screenshots and turn pages into Markdown on Cloudflare's browsers Sign in

Frameworks

Connector (id) What it gives Albert Asks for
Next.js DevTools (next-devtools) Next.js documentation, upgrades and a running app's errors and routes
Angular CLI (angular-cli) Angular best practices, documentation search and the projects in an Angular workspace
Nx (nx) An Nx workspace's projects, tasks and generators, and Nx documentation The workspace folder
Shopify Dev (shopify-dev) Shopify's developer documentation and API schemas, and checks for Liquid and GraphQL
ESLint (eslint) Lint JavaScript and TypeScript files with the project's own ESLint setup
n8n (n8n) Every n8n node's documentation and properties, and checks for workflows before you import them

Cloud and deployment

Connector (id) What it gives Albert Asks for
Vercel (vercel) Projects, deployments, build logs and documentation on Vercel Sign in
Netlify (netlify) Create, deploy and manage Netlify sites, forms and environment variables Sign in
Cloudflare Workers (cloudflare-bindings) Workers, KV, R2, D1 and Hyperdrive in your Cloudflare account Sign in
Cloudflare Workers Builds (cloudflare-builds) Builds and their logs for your Cloudflare Workers Sign in
Cloudflare Observability (cloudflare-observability) Your Workers' logs and analytics, searched and summarised Sign in
Cloudflare Radar (cloudflare-radar) Internet traffic trends, outages, attacks and URL scans from Cloudflare Radar Sign in
Railway (railway) Projects, services, deployments, logs and variables on Railway Sign in
CircleCI (circleci) Pipelines, failed jobs, their logs and flaky tests on CircleCI Sign in
Buildkite (buildkite) Pipelines, builds, jobs and their logs on Buildkite Sign in
Docker (docker) Containers, images, networks and volumes on this machine's Docker Needs Docker running
Kubernetes (kubernetes) Pods, deployments, services, logs and Helm charts in the clusters your kubeconfig names Needs kubectl

Databases

Connector (id) What it gives Albert Asks for
PostgreSQL (postgres) Read-only SQL, schema details and query plans for one PostgreSQL database The connection URI (uri)
SQLite (sqlite) Query and change one SQLite database file, and describe its tables The database file
DuckDB (duckdb) Read-only SQL over a DuckDB database, and over the CSV, Parquet and JSON files it can read The database file
MySQL (mysql) SQL and table listings for one MySQL or MariaDB database Host, port, user, password, database
MongoDB (mongodb) Collections, documents, indexes and aggregations in MongoDB, read-only The connection string (uri)
Redis (redis) Keys, hashes, lists, streams, JSON and vector search in a Redis database The Redis URL (url)
Neo4j (neo4j) The graph's schema, and Cypher queries against a Neo4j database URI, user, password
ClickHouse (clickhouse) Databases, tables and read-only SQL on a ClickHouse server Host, port, user, password
Chroma (chroma) Collections and documents in a Chroma vector database kept in a folder The data folder
Supabase (supabase) Tables, SQL, migrations, edge functions and logs in your Supabase projects Sign in
Neon (neon) Projects, branches, SQL and migrations on Neon serverless Postgres Sign in
Prisma Postgres (prisma) Create and manage Prisma Postgres databases and run migrations Sign in
PlanetScale (planetscale) Databases, branches, schema, query insights and SQL on PlanetScale Sign in

Monitoring and analytics

Connector (id) What it gives Albert Asks for
Sentry (sentry) Errors, issues, traces and releases in Sentry, and Seer's suggested fixes Sign in
PostHog (posthog) Product analytics, feature flags, experiments and errors in PostHog Sign in
Amplitude (amplitude) Charts, dashboards, cohorts and experiments in Amplitude Sign in
Mixpanel (mixpanel) Events, funnels, retention and reports in Mixpanel Sign in
Honeycomb (honeycomb) Traces, queries, SLOs, triggers and boards in Honeycomb Sign in
Axiom (axiom) Datasets, schemas and APL queries over your logs and events in Axiom Sign in
Datadog (datadog) Logs, metrics, traces, monitors, incidents and dashboards in Datadog Sign in
New Relic (newrelic) Metrics, logs, traces and alerts in New Relic, queried in plain language or NRQL Sign in

Design and content

Connector (id) What it gives Albert Asks for
Canva (canva) Create, search, edit and export Canva designs Sign in
Miro (miro) Read and add to Miro boards: diagrams, documents, tables and sticky notes Sign in
Lucid (lucid) Create, search and summarise Lucidchart and Lucidspark documents Sign in
Whimsical (whimsical) Flowcharts, mind maps, wireframes and docs in Whimsical Sign in
Webflow (webflow) Sites, pages, CMS collections and items in Webflow Sign in
Wix (wix) Build and manage Wix sites, and search Wix's developer documentation Sign in
Sanity (sanity) Documents, datasets, releases and schemas in your Sanity projects Sign in

Business and payments

Connector (id) What it gives Albert Asks for
Stripe (stripe) Customers, products, prices, payments and refunds in Stripe, and Stripe's documentation Sign in
PayPal (paypal) Invoices, orders, payments, disputes and subscriptions on PayPal Sign in
Square (square) Payments, orders, catalog, customers and inventory on Square Sign in
Intercom (intercom) Conversations, contacts and help-centre articles in Intercom Sign in
Close (close) Leads, contacts, opportunities and activities in Close CRM Sign in
Attio (attio) Records, lists, notes and tasks in your Attio CRM Sign in
PandaDoc (pandadoc) Create, send and track PandaDoc documents and templates Sign in
Apify (apify) Run thousands of ready-made web scrapers and automations from the Apify Store Sign in
WorkOS (workos) Organisations, users, SSO, Directory Sync and AuthKit in WorkOS Sign in

Utilities

Connector (id) What it gives Albert Asks for
Memory (memory) A knowledge graph the agent can keep notes in across sessions
Sequential Thinking (sequential-thinking) A place to work through a hard problem step by step
Time (time) Current time and time-zone conversion
Excel (excel) Read, write and format Excel workbooks: sheets, formulas, charts and pivot tables, without Excel
MarkItDown (markitdown) Turn PDF, Word, Excel, PowerPoint, HTML and other files into Markdown the agent can read
Everything (test server) (everything) The protocol's reference server: sample tools, prompts and resources, for trying MCP out

Local connectors need what they run on: uv (uvx), Node.js (npx), and for some Chrome, Docker or kubectl. The gallery says which. The first start downloads the server, so it may take a minute; later starts are quick.

Interface How
Browser MCP servers… in the Settings menu (or /mcp): under Ready to add, pick a category chip or search, then Add. One that needs an answer (a key, a folder) asks for it, with a Server name of your choosing; one that signs in opens its sign-in page in a new tab as it is added
Terminal /mcp, then + Add a connector…
Classic terminal /connectors, or /connectors postgres to search; /mcp install <id> key=value adds one without questions
Classic page MCP servers (⋮ menu) → Browse connectors, or Connectors in the command palette
VS Code Browse MCP Connectors
/mcp install postgres uri=postgresql://localhost:5432/app
/mcp install git repository=.
/mcp install mongodb uri={env:MONGODB_URI}
/mcp install linear              # then sign in: your browser opens Linear's own sign-in page
/mcp install context7            # no sign-in: up-to-date library docs straight away

A connector that signs in, such as Linear, Notion, Jira, Stripe or Sentry, needs nothing but your account: AlbertCode registers itself with the service, opens its sign-in page, and comes back connected once you approve. For example, to give Albert your Linear issues:

  1. Browse connectors → Projects and work → Linear → Connect.
  2. Linear's own page opens in a new tab: sign in, and approve AlbertCode.
  3. Back in AlbertCode, Linear shows Connected, with its tools. Ask "What is assigned to me this week?"

A few popular services are not in the gallery yet, because the check could not show them working: Box, HubSpot, ElevenLabs and Render make you register a client by hand; Figma and Sonatype take only clients they approved; GitHub needs your personal access token to be checked. Any of them can still be added by hand: Add server, with its URL (and a client ID where it needs one).

Add any server

Interface How
Terminal /mcp add, then answer: a local command, or a remote URL
Browser MCP servers → Add server: a name, then a Local command (its command and environment) or a Remote URL (its address, OAuth or headers)
VS Code Add an MCP Server
A script albertcode mcp add … (below)
Terminal
# A local server: a command AlbertCode starts, with its variables
albertcode mcp add docs --command "npx -y @example/docs-mcp" --env DOCS_TOKEN=xyz

# A remote server that signs you in with OAuth (a browser window opens)
albertcode mcp add tracker --url https://mcp.example.com/mcp

# A remote server that takes a header instead
albertcode mcp add search --url https://search.example.com/mcp --no-oauth --header "Authorization=Bearer xyz"

Signing in

A remote server that needs a sign-in shows Needs auth. /mcp auth <name> (or Sign in in the browser and VS Code) opens your browser to sign in; the token is kept in your system's key store and refreshed for you. /mcp logout <name> (Sign out in the browser) removes it. A server that needs a client ID registered by hand shows Needs client ID: give it as the server's OAuth client ID.

The sign-in asks only for what the server says it needs. GitLab's MCP server, for example, asks for the mcp permission alone, so that is all the sign-in page asks you to grant, even though GitLab's sign-in offers many more (administrator rights among them). To ask for something else, give the server a scope of your own in mcp.json, the file beside your credentials that holds your servers:

JSON
"tracker": {"type": "remote", "url": "https://mcp.example.com/mcp", "oauth": {"scope": "read write"}}

Managing servers

Terminal Browser (MCP servers, then a server's Edit) What it does
/mcp The list Pick one to switch it on or off
/mcp list The list Every server, its status and its tools
/mcp tools <name> Its tool switches Switch single tools on or off
/mcp enable <name>, /mcp disable <name> Its switch Switch a server on or off
/mcp connect <name> Reconnect Connect again
/mcp debug <name> Its details Why a server is not working: its command, its output, the error
/mcp remove <name> Delete Remove a server

In VS Code, MCP Servers offers the same.

Status Meaning
Connected · 12 tools · 3 prompts Working, with what it offers
Connecting… Starting or reaching it
Needs auth Sign in with /mcp auth <name>
Needs client ID The server needs a client registered by hand
Disabled Switched off
Failed: … With the error; /mcp debug <name> says more

The sidebar's MCP chip in the browser shows how many are connected, and the terminal says when a server connects, drops or wants a sign-in. A server that drops is reconnected by itself.

Prompts and resources

Servers can offer prompts (ready-made requests) and resources (documents and data):

/mcp prompts                       # what servers offer
/github:review-pr 128              # use a prompt: /server:prompt arguments
/mcp resources                     # what servers offer
/mcp attach docs docs://api/orders # attach a resource to your next message

In the browser, MCP prompts and resources in the command palette; in VS Code, Use an MCP Prompt or Resource. A prompt's text is sent as your message, shown in full.

What a server may do

An MCP server's tools are governed like everything else:

  • While Albert plans or answers questions, only tools that read are offered.
  • A tool that acts (writes, sends, changes something) needs your approval: Allow in the browser and VS Code (or Allow, and stop asking), a in the terminal.
  • Every call is recorded in the evidence ledger.
  • A custom agent without the mcp group cannot use MCP tools at all.

Servers you add are yours, for every repository. Servers a plugin brings are named <plugin>-<name> and removed with the plugin.

From docs/mcp.md in huynvic/albertcode-swe-agent at a8cef99 (15 Oct 2026), shown unchanged except for links. AlbertCode SWE Agent documentation, CC BY 4.0. Latest on GitHub.