Add Mida plugin

.cursor-plugin/marketplace.json

@@ -307,6 +307,11 @@
       "name": "xero",
       "source": "third_party/xero",
       "description": "Read and write invoices, contacts, reports, and payroll."
+    },
+    {
+      "name": "mida-so",
+      "source": "third_party/mida-so",
+      "description": "Create A/B tests, check results, and roll out winners on your site."
     }
   ]
 }

README.md

@@ -66,6 +66,7 @@ Official Cursor plugins for popular developer tools, frameworks, and SaaS produc
 | `readwise` | [Readwise](third_party/readwise/) | Cursor | Integrations | Search highlights and Reader documents, save articles. |
 | `similarweb` | [Similarweb](third_party/similarweb/) | Cursor | Integrations | Analyze website traffic, audiences, and competitors. |
 | `xero` | [Xero](third_party/xero/) | Cursor | Integrations | Read and write invoices, contacts, reports, and payroll. |
+| `mida-so` | [Mida](third_party/mida-so/) | Mida | Integrations | Create A/B tests, check results, and roll out winners on your site. |
 Author values match each plugin’s `plugin.json` `author.name` (Cursor lists `plugins@cursor.com` in the manifest).
 
 ## Repository structure

third_party/mida-so/.cursor-plugin/plugin.json

@@ -0,0 +1,30 @@
+{
+  "name": "mida-so",
+  "displayName": "Mida",
+  "version": "1.0.2",
+  "description": "Create A/B tests, check results, and roll out winners on your site.",
+  "author": {
+    "name": "Mida"
+  },
+  "homepage": "https://www.mida.so/mcp",
+  "repository": "https://github.com/cursor/plugins",
+  "license": "MIT",
+  "logo": "assets/logo.svg",
+  "keywords": [
+    "mida",
+    "ab-testing",
+    "experimentation",
+    "personalization",
+    "cro",
+    "mcp"
+  ],
+  "category": "integrations",
+  "tags": [
+    "mida",
+    "ab-testing",
+    "experimentation",
+    "analytics"
+  ],
+  "skills": "./skills/",
+  "mcpServers": "./mcp.json"
+}

third_party/mida-so/CHANGELOG.md

@@ -0,0 +1,10 @@
+# Changelog
+
+## 1.0.2
+
+- Renamed the plugin to `mida-so`, matching the GitHub org and repo prefix.
+
+## 1.0.0
+
+- Initial release: Mida MCP server (hosted, OAuth) plus the `mida` and
+  `mida-experiments` skills.

third_party/mida-so/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Mida
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

third_party/mida-so/README.md

@@ -0,0 +1,72 @@
+# Mida
+
+**Test everything. Personalize everyone.**
+
+Run A/B tests and personalizations on your own website, from Cursor.
+
+[Mida](https://www.mida.so) is an A/B testing and personalization platform. A
+small JavaScript tag on your site puts visitors into variants and reports
+conversions. This plugin connects Cursor to your Mida account, so you can build
+a test, launch it, read the result, conclude it, and roll the winner out while
+you're still in the code the test is changing.
+
+```
+"What's running on the pricing page right now?"
+"Create a test for a shorter signup form and preview variant B"
+"Has the homepage hero test reached significance yet?"
+"Conclude experiment #208, variant B won, we're shipping it"
+```
+
+## Setup
+
+On first use, Cursor opens a browser to sign in to Mida with OAuth. There is no
+API key to paste and nothing is written to your machine.
+
+You need a [Mida account](https://www.mida.so) and the Mida tag installed on the
+site you want to test. Ask for the install snippet once you're connected.
+
+## What's included
+
+**MCP server.** The hosted Mida MCP server, covering experiments (create,
+launch, update, conclude, reopen), results and statistics, goals and custom
+events, hypotheses, exclusion groups, personalization campaigns, project
+configuration, and team access.
+
+**Skills**
+
+| Skill | Covers |
+|---|---|
+| `mida` | Entry point: picking the right project, reading Bayesian vs. frequentist results, and the order of operations for concluding a test and serving a winner |
+| `mida-experiments` | The workflow: hypothesis, variants, goal, preview, launch, then calling the result |
+
+The skills are here because a few Mida behaviours are easy to get wrong from the
+tool schemas alone. An experiment's display number is not its API id. A
+concluded test stops serving even while its status still reads active. Setting a
+test's status clears a conclusion written before it. Each one is documented with
+the correct sequence.
+
+## Network and credentials
+
+| | |
+|---|---|
+| **Endpoints called** | `https://mcp.mida.so/mcp`, the hosted Mida MCP server, and the only host this plugin contacts. It reaches Mida's API server-side. |
+| **Credentials** | OAuth to your Mida account, handled by the MCP client. No API keys, no environment variables, no local credential files. |
+| **Scope** | Your own Mida organizations and projects. The plugin reads and writes experiment configuration and reads result data for those projects. |
+| **Local execution** | None. No hooks, no commands, no scripts. One remote MCP server plus Markdown skills. |
+
+## About Mida
+
+[Mida](https://www.mida.so) is A/B testing and personalization in one engine.
+Testing answers which version works. Personalization answers who it works for.
+Both run as real variants in a real test, so personalization is something you
+can actually prove rather than take on faith.
+
+- [Mida](https://www.mida.so), A/B testing on any site, with a tag that loads in about 20ms
+- [Mida dashboard](https://app.mida.so) for results, goals, and rollouts
+- [Public API docs](https://github.com/mida-so/api-docs)
+- [MCP setup guide](https://www.mida.so/mcp) for every client
+- [Plugin source](https://github.com/mida-so/mida-plugin)
+
+## License
+
+MIT, see [LICENSE](LICENSE).

third_party/mida-so/mcp.json

@@ -0,0 +1,8 @@
+{
+  "mcpServers": {
+    "mida": {
+      "type": "http",
+      "url": "https://mcp.mida.so/mcp"
+    }
+  }
+}

third_party/mida-so/skills/mida-experiments/SKILL.md

@@ -0,0 +1,98 @@
+---
+name: mida-experiments
+description: >-
+  Build and launch a Mida A/B test end to end: writing the hypothesis, creating
+  the experiment and its variants, attaching a goal, previewing the variant
+  before it goes live, launching, and deciding when there is enough data to
+  call it. Use when the user asks to test, try, or experiment with a change to
+  their site ("test a new headline", "try a different CTA", "does the shorter
+  form convert better"), or to review a test someone has already set up.
+---
+
+# Running an experiment in Mida
+
+Read the `mida` skill first for project selection and result-reading rules. This
+skill is the workflow.
+
+## 1. Start from a hypothesis, not a change
+
+A test is only worth running if you can say in advance what you expect and how
+you would know. Push for the shape:
+
+> Changing **X** for **audience Y** will improve **metric Z**, because **reason**.
+
+`create_hypothesis` stores it, and `link_hypothesis_to_experiment` ties it to the
+test so the conclusion can be read against what was actually predicted. Before
+writing a new one, `find_similar_tests` and `get_my_testing_history` will show
+whether this account has already run something close. A repeat of a test that
+already lost is the most common waste.
+
+## 2. Change one thing
+
+If a variant changes the headline, the button colour, and the form length at
+once, a win tells you nothing about which of the three did it, and a loss tells
+you nothing about which to keep. When a user describes a bundle of changes, say
+so and offer to split it, unless they are deliberately testing a whole new
+page, which is a legitimate and different kind of test.
+
+Keep the control as-is. Never "improve" the control while building the variant.
+
+## 3. Create the experiment
+
+`create_experiment` takes the target URL rule, the variants, and the traffic
+split. Two things to get right at creation time:
+
+- **The URL rule is a pattern.** A test meant for every product page needs a
+  rule that matches them all; a rule pasted from one specific URL will match
+  that page only. Confirm which the user means.
+- **Query strings are matched literally.** An entry URL written with `?param=x`
+  will only match visitors whose URL carries that exact query string, which
+  usually is not what someone wants.
+
+`get_experiment_distribution` after launch confirms traffic is actually splitting
+the way you configured it.
+
+## 4. Attach the goal before launching
+
+An experiment with no goal collects visitors and measures nothing. Check
+`list_goals` for an existing goal that already tracks the action, and reuse it
+rather than creating a near-duplicate. Duplicates split the same conversions
+across two metrics.
+
+Pick a primary goal that the change could plausibly move. A headline test
+measured only on checkout completion will almost never reach significance;
+measure the click the headline is trying to earn, and keep the downstream metric
+as a secondary via `update_experiment`.
+
+## 5. Preview before you launch
+
+`generate_preview_url` returns a link that forces a given variant, so the user
+can see the change on their real site before a single visitor does. Always offer
+it. Broken variant markup that ships to 50% of traffic is the expensive failure
+mode here, and it is entirely avoidable.
+
+## 6. Launch and leave it alone
+
+`update_experiment_status` starts delivery. Then:
+
+- **Do not read the result on day one.** Early numbers swing hard and reading
+  them invites stopping on noise.
+- **Run at least one full business cycle**, usually a week, so every weekday is
+  represented in both arms. A test stopped on Wednesday inherits Wednesday's
+  traffic mix.
+- **Do not edit a running test.** Changing the variant, the goal, or the
+  targeting mid-flight mixes two different experiments into one dataset. If the
+  change is necessary, stop the test and start a fresh one.
+
+## 7. Call it
+
+Use `get_experiment_result` with `compute_experiment_statistics`, and check
+`get_experiment_timeseries` to see whether the result has settled or is still
+moving. Then `conclude_experiment` with a written conclusion: what changed,
+what happened, and what to do next. That write-up is what makes the account's
+history worth anything later; a concluded test with an empty conclusion is a
+test nobody can learn from.
+
+If the winner should keep serving while the change gets built for real, use
+`serve_experiment_winner`, and see the ordering rules in the `mida` skill
+before deploying that variant to source.