Claude's connector directory crossed 200+ one-click integrations in April 2026, up from 50 in February
I tested 12 connectors over a weekend and replaced five standalone scripts I had been maintaining for months
The directory works on every Claude plan, including Free, which makes it the lowest-friction MCP entry point on the market
Linear, GitHub, Notion, Gmail, and Stripe were the five connectors that earned a permanent spot in my workflow
Three connectors (Spotify, Uber, TripAdvisor) felt like demo-ware, fine for a tweet, useless for daily work
MCP is the open protocol underneath, which is exactly why this scales where ChatGPT plugins stalled out
I spent a weekend installing 12 Claude connectors and watching my normal workflow dissolve into chat threads. The directory hit 200+ integrations this month, up from around 50 in February. Here is what actually changed, what I kept, and what I uninstalled by Sunday night.
What Changed When Claude Started Doing Things In My Tools
For two years I treated Claude like a very smart text box. I pasted things in. I copied things out. The loop was: ask Claude, get answer, switch tab, do work, come back. Connectors broke that loop.
The first time it clicked was a Friday afternoon. I told Claude "show me my open Linear issues tagged urgent and group them by which ones block a deploy." Forty seconds later I had a prioritized list inside the chat, with links back to each issue. Total elapsed time, including auth, was under a minute. The same task in the Linear UI, with my own filters and saved views, takes me about twenty minutes because I get distracted reading comments.
That gap (one minute vs twenty) is the whole pitch. It is not that Claude is faster than Linear. It is that Claude does not ask me to context-switch into Linear's UI, with its sidebars and notifications and Slack pings I see along the way. The work happens where the thinking happens.
I noticed three behavior shifts after the first day:
I stopped opening Notion. I just asked Claude to fetch the page I needed.
I started ending threads with "and now post this to the GitHub issue" instead of copy-pasting.
I trusted Claude with read-only operations way faster than write-back ones.
The third point matters. Reading from a tool feels safe. Writing back feels like handing over the keys. I will get to where the line lands later in this post, because it is the single biggest thing that determines whether a connector earns a permanent install.
The bigger frame here: AI tools spent 2024 and 2025 racing to know more things. Bigger context windows, fresher knowledge, better retrieval. Connectors are the first signal that the next race is doing more things in the tools you already pay for. Knowing is cheap now. Acting is the moat.
The shift also changes how I think about subscriptions. I had been quietly auditing my SaaS spend, asking which tools earn their monthly seat. With connectors, the calculus moves. A tool I touch three times a week through Claude is suddenly more valuable than a tool I open daily but barely use. Linear got promoted on my mental list. A few read-mostly dashboards got demoted, because Claude can summarize them on demand and I do not need to log in.
The 200+ Connector Sprawl, Categorized
The directory at claude.ai/directory/connectors is a wall of tiles, sorted by category. After scrolling through twice, here is the rough shape of it:
Communication. Slack, Gmail, Microsoft 365. The three you would expect. Slack is the one I install first on every new Claude account because asking "what did I miss in #engineering today" is a daily-use question.
Project management. Asana, Linear, Jira, Monday.com. Linear is the standout for solo and small-team work. Jira is necessary if you are stuck with it but the connector is honest about being a wrapper.
Content. Notion, Google Drive, WordPress.com. Notion is the one most people will reach for. Google Drive is quietly useful because it indexes Docs, Sheets, and Slides in one connector.
Design. Canva, Figma. Figma is read-mostly right now, which is fine. I do not want Claude rearranging my frames.
Engineering and data. GitHub, Hex, Amplitude. GitHub is the workhorse. Hex and Amplitude matter if you live in dashboards.
Finance. Stripe. One connector, very high value.
Healthcare. Apple Health, PubMed. Apple Health is a curiosity for most people and a daily tool for some. PubMed is a research multiplier.
Lifestyle. Spotify, Instacart, Uber, TripAdvisor. This is where I rolled my eyes a little.
The growth curve is the part worth pausing on. July 2025: launch. February 2026: 50+ curated integrations. April 2026: 200+, growing weekly. That is roughly 4x in two months. The directory page tells you which ones are official (built by Anthropic or the partner) and which are community-built. The split matters because community connectors can be excellent or abandoned within the same week.
I install official connectors without thinking. Community ones I check the GitHub last-commit date before connecting OAuth.
A useful filter when browsing: ignore anything that does not have a write-back path. A read-only connector is a search bar with extra steps. A connector that can read context, draft an action, and let me approve and send the action is where the time savings live.
One thing the directory does well is surface the auth scopes upfront. Before I click connect, I see exactly what permissions the connector wants. Most ask for the minimum (read messages, draft items, post to a single workspace). A few ask for blanket scopes that I declined on principle, even when the app itself was useful. If a connector wants full mailbox write access just to draft replies, that is a no from me. I would rather lose the feature than hand over send authority.
Five Connectors That Replaced Standalone Workflows
These five earned a permanent install. I rank them by how often I actually use them, not how impressive the demo is.
1. Linear. Used daily. I pull open issues, prioritize them in chat, draft new ones from voice notes, and let Claude write the rephrased version into Linear after I approve. Saved me about 40 minutes a day on issue grooming, mostly because I stopped re-reading old comment threads I did not need.
2. GitHub. Used daily. Reviewing PRs in chat is a different rhythm than reviewing them in the GitHub UI. I get a structured summary, ask follow-ups, and then post the actual review comment back. The review comment quality is noticeably better, because I am not skim-reading code, I am responding to a synthesis. This is also where I lean on Git Dojo when teammates need to level up their git mental model before they touch the review queue.
3. Notion. Used weekly. Pulling specs into a Claude chat for technical questions, then asking Claude to draft an updated section. Write-back works but I always preview the diff before approving. The connector is fast, which is more than I can say for the Notion app on my laptop.
4. Gmail. Used a few times a day. The killer use case is "summarize the threads I missed and flag anything that needs a response by tonight." I do not let Claude send mail on my behalf yet. Drafts only. The reply quality is fine, but the trust gap on outbound communication is real and I am keeping the human in the loop.
5. Stripe. Used weekly for the studio. Pulling subscription health, dunning data, and payment status into a chat is a different experience than navigating the Stripe dashboard. The dashboard is built for finance teams. The chat is built for a founder asking one specific question. If you run a Shopify storefront alongside a SaaS layer, having both Stripe and Shopify connectors in the same chat is the closest thing to a unified ops view I have seen.
The pattern across all five: high-frequency tasks, structured data, and a clear write-back boundary I control. None of these are flashy. All of them save me real minutes every day.
Three Connectors That Sound Useful But Aren't
I want to be honest about the ones I uninstalled. Not because they are bad, but because the use case did not survive contact with my actual week.
1. Spotify. Cool to demo. "Make me a playlist for deep work." I did this twice, smiled, and never used it again. The playlists were fine. The friction of asking Claude to do it instead of opening Spotify's own AI playlist feature was negative. Spotify's native UI is faster for this exact task. The connector is solving a problem I did not have.
2. Uber. Same story. "Book me a ride home" is one tap in the Uber app. Adding a chat layer makes it slower, not faster. I can see a future version where Claude orchestrates a trip (book flight, book hotel, book ride, calendar it all) but the current single-action connector does not unlock that. It is a tile on the directory more than a workflow.
3. TripAdvisor. Travel research is something I do twice a year. The connector pulls reviews and ratings, which is fine, but I am already going to spend an hour on this and the ten minutes saved by a connector does not change my behavior. Compare that to Linear, where small minutes compound across hundreds of touches per week. The math just is not there for low-frequency tasks.
The honest pattern: connectors win when the task is high-frequency, lives in structured data, and benefits from synthesis across multiple records. Connectors lose when the native app is already optimized for one-tap actions. Spotify, Uber, and most lifestyle apps fall in the second bucket.
There is a related failure mode: OAuth fatigue. By connector number nine I was tired of clicking through scope screens. Anthropic could help here by grouping related connectors into bundles (a "Google bundle" or a "developer bundle") so I auth once for a related set. Right now each one is its own ceremony.
One more honest call-out: connector quality varies a lot. Official partner-built connectors (Linear, GitHub, Stripe) are tight. Some community connectors I tried had clunky parameter handling, ambiguous error messages, or returned data that needed reshaping before it was useful. Read the description. Check who built it.
MCP Is The Reason This Works (And Why ChatGPT Plugins Didn't)
The interesting part is not the directory. It is what is under the directory.
Connectors are built on MCP, the Model Context Protocol that Anthropic open-sourced in late 2024. MCP is a specification for how an AI model talks to external tools: how it asks for permissions, how it describes capabilities, how it handles errors. Crucially, it is open. Anyone can build an MCP server. Anyone can host one. The directory is just the consumer-facing storefront for a much bigger ecosystem.
This is the part that quietly leapfrogged ChatGPT plugins.
ChatGPT plugins launched in March 2023 with huge fanfare and were effectively dead within a year. The reasons were structural: plugins were a closed system, every integration had to be approved through OpenAI, the install flow was clunky, and developers had no path to host their own. Once excitement faded, the directory rotted.
MCP solves all four problems. It is an open spec, so any client (Claude, IDEs, third-party tools) can implement it. Builders can self-host or partner-host their MCP server, no gatekeeper required. The Claude directory is one front-end onto MCP, not the only one. And because the spec is public, the same MCP server I built for myself this morning works in any MCP-compatible client tomorrow.
The economics work because Anthropic does not need to monetize connectors. Connectors are a feature that makes Claude stickier. The MCP servers are owned by the partner companies. Linear runs Linear's MCP server. Stripe runs Stripe's. That alignment is healthy. Nobody is fighting over a 30 percent platform tax.
For solo operators and small teams, this is a quiet upgrade. I am one person. I do not have engineering resources to integrate fifteen SaaS tools. The directory hands me those integrations for free, on every plan, including Claude Free. That is the part that surprised me most when I dug into the pricing page. Connectors are not an enterprise upsell. They are the table stakes.
For enterprises, the calculus is different. Most large companies will want to host their own MCP servers internally, with their own auth and audit logs, pointing at internal systems. The open spec lets them do that without vendor lock-in. The same Claude client connects to the public directory and to the company's internal MCP fleet. That is also why I think the directory will keep growing. It is not just a marketing surface. It is the on-ramp for an ecosystem that is now bigger than Claude itself.
The other quiet win is portability. Because MCP is a spec, the work I do today connecting Linear, GitHub, and Stripe is not Claude-specific. If a competitor ships a better client tomorrow, the same MCP servers point at it. That removes the lock-in fear that kept me from going all-in on ChatGPT plugins three years ago. I am not building on a platform that can deprecate my workflow. I am building on a protocol that any client can implement.
If you want to go deeper into how MCP servers actually get built, I wrote a separate breakdown in the Lab and the Claude Blueprint goes into the production patterns I use day to day.
Bottom Line
The 200+ connector directory is not a flashy launch. It is a quiet shift in what AI assistants are for.
Twelve months ago I asked Claude to know things. Today I ask it to do things in tools I already pay for, and it does them with a one-click OAuth and a chat thread instead of a custom integration project. That changes which problems are worth solving with code vs. just asking the assistant.
If you are starting from zero, install five and only five: Linear (or your project manager), GitHub (or your code host), Gmail, Notion (or your docs tool), and your billing system. Use them for a week. Add others only when a real workflow demands it. Skip the lifestyle connectors unless you find a daily-frequency use case. Keep write-back permissions tight until you trust each connector individually.
The bigger move: track how often you reach for a tool's native UI. If a connector replaces three or more daily UI visits, it stays. If not, uninstall it. The directory will keep growing, and the discipline of pruning matters more than the discipline of installing.
If you want to see how I wire connectors into actual studio workflows, studio.raxxo.shop shows the live setup, and the writeups land in the Lab every week. For social scheduling on top of all this, Buffer is still my pick because it stays out of the way of the chat-first workflow Claude is becoming.
Top comments (0)