# MCP Release Checklist

Alkenacode, version 1.0.0, 2026-09-28.
Free to use and adapt in your projects. Provided as a review aid, without a warranty or client compatibility certification.

## Define the release

- [ ] State what the server does, supported environments, and required credentials.
- [ ] Pin SDK and protocol versions and link their matching documentation.
- [ ] Select stdio for supported local clients or the remote transport your client accepts.
- [ ] Document runtime, installation, configuration, start, upgrade, and uninstall steps.

## Verify the contract

- [ ] List tools successfully in each supported client.
- [ ] Run each tool with valid, missing, malformed, and boundary-value inputs.
- [ ] Verify result schemas and actionable errors.
- [ ] Verify denied access and expired or revoked credentials.
- [ ] Restrict each token to the site and operations it needs.
- [ ] Keep secrets and sensitive user data out of errors and logs.

## Verify changes and recovery

- [ ] Preview the exact draft revision before a publish action.
- [ ] Reject approval for a stale or changed revision.
- [ ] Enforce permissions on the server for every write.
- [ ] Retry a write and confirm it does not create a second change.
- [ ] Exercise timeouts, disconnects, cancellation, and rate limits.
- [ ] Record an audit event and test the documented rollback or correction flow.

## Record client compatibility

| Client and version | Account/region | Transport | Auth | Server version | Date | Result/evidence |
| --- | --- | --- | --- | --- | --- | --- |

A mock test does not certify a real client. Repeat real-client checks when relevant dependencies change.

## Measure what the package saves

Use the same task and completion criteria for both runs. Include failed attempts and setup time.

| Run | Model/version | Input tokens | Cached input | Output tokens | API charge | Active time | Elapsed time | Acceptance result |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |

API cost = ((input - cached input) * input rate + cached input * cached rate + output * output rate) / 1,000,000.

Net first-use value = API cost avoided + hours avoided * chosen hourly value - asset price - additional setup/hosting cost.

Record actual usage. A context-window limit is not a token count. Label estimates, disclose sample size, and report failures.

## References

- MCP specifications: https://modelcontextprotocol.io/specification
- ChatGPT MCP documentation: https://developers.openai.com/api/docs/guides/tools-connectors-mcp
- Gemini custom apps: https://support.google.com/gemini/answer/17209137

More tools and implementation help: https://alkenacode.dev/shop