# Firenook 0.2.0-next.7 release notes > Firenook 0.2.0-next.7, 20 October 2026: a project with no seed keeps its data, --keep-data, firenook mcp add, 54 MCP tools, up from 19, and more. Data kept without a seed, coding agents set up in one command, 54 MCP tools. Published on 20 October 2026, as a preview. `npm install --save-dev firenook@0.2.0-next.7` [Every release](https://firenook.dev/changelog) A newer release is out: [Firenook 0.2.0](https://firenook.dev/changelog/0.2.0), the first release for public use. ## What changed 11 changes you will notice, as the release notes give them. 1. ### A project with no seed keeps its data `--resume-state` needed `--import`, so a project made with `firenook init` could not keep its data between runs. Now it needs only a state directory; data started without a seed records an empty one, and a start that mixes the two up (a seedless state given `--import`, or a seeded one started without it) is refused with the way on, nothing changed. 2. ### `--keep-data` is short for `--state-dir ./.firenook/state --resume-state`, with or without `--import`. Keeping data stays opt-in, so `--import ./data` alone still starts from that data every time. `init` prints its start command with `--keep-data`. 3. ### `firenook mcp add` gives the project's coding agents the MCP server: it writes `.mcp.json` (Claude Code), `.cursor/mcp.json` and `.vscode/mcp.json` for the agents it finds, keeping every other server, with `--dry-run`, `--agent` and `--force`. A second run changes nothing. `init` offers it once in a terminal; `init --adopt` lists it in its plan. The terminal's `?` help names it. 4. ### 54 MCP tools, up from 19 The function calls and the log (with each call's error and stack trace, and a replay), Security Rules (the rules in force, whether a request as a given user is allowed and which line decided, recent denials), Auth tokens and claims, Storage files (read, upload, delete, copy, move), Eventarc, Extensions, Pub/Sub topics and a backlog peek, Cloud Tasks, schedules, Firestore counts and collection-group queries, and a reset that asks for the project id before it clears more than one service. The new tools use the console's API, so they need the console on (the default). A `{"$timestamp"}` finer than a millisecond now keeps its precision when a tool writes it. 5. ### `firenook mcp` with no arguments uses the one running suite, so an editor's install link needs no project id; with two running, it asks for `--project`. 6. ### The console's Connect dialog has a Coding agent tab : the setup for Claude Code, Cursor, VS Code and any other agent, "Add to Cursor" and "Add to VS Code" install links, and when an agent last called. 7. ### Functions start in under a second, not seven Discovery waited about 6 s for the SDK's discovery server to exit, kept open by an idle keep-alive connection; it now takes about 0.2 s. On the acceptance host, a project with seven functions and every emulator is ready in 0.78 s against 6.82 s. 8. ### Timestamps are stored to the microsecond , as in production: a finer value is rounded down at write, so the export no longer refuses it, and the export rounds down any finer value an earlier release saved. 9. ### The Functions log line's duration prints `1.6ms`, not `1.6076249999999999ms`. 10. ### The console's document panel reads a bare id in `doc=` against the open collection instead of crashing. 11. ### Checked against 0.2.0-next.6 on the acceptance host with the same data, instead of a new paired run against the official emulator, since no measured path changed: imports, resumed starts, writes, listener delivery and the query shapes match within run-to-run noise, and memory is 11 MB higher once ready (`docs/benchmarks.md`, `benchmarks/next7-regression.json`). The slow start the 0.2.0-next.6 run saw on a freshly copied state directory is the copy's pending writes reaching the disk, not Firenook. ## What got faster Measured on the same data with 0.2.0-next.6 and with 0.2.0-next.7. Lower is better. ### Ready, with seven functions and every emulator 8× sooner - [0.2.0-next.6](https://firenook.dev/changelog/0.2.0-next.6) 6.82 s - [0.2.0-next.7](https://firenook.dev/changelog/0.2.0-next.7) 0.78 s Every multiple is rounded down. [Every release’s measures](https://firenook.dev/changelog#faster). ## Your data What this release does to data an earlier one kept, and to the ports and protocols. - Data kept without a seed needs 0.2.0-next.7; 0.2.0-next.6 refuses it and changes nothing. Data saved by 0.2.0-next.6 opens unchanged, and data 0.2.0-next.7 saved from a seed opens in 0.2.0-next.6 (checked on the private consumer seed: 211,202 documents and 33,353 objects). No emulator protocol or port changed. ## How it was checked The source it was built from, the runs it passed, and how it was published. - Pin the engine to `37fdd0387ea81f3131c20b594159113c86084a3a`, the merge of [PR #107](https://github.com/firenook-dev/firenook/pull/107) on 0.2.0-next.6. Its tree is the PR head's, `e8ecd33`, which passed CI ([run 39450933562](https://github.com/firenook-dev/firenook/actions/runs/39450933562)) and the five-platform packed-install matrix ([run 39450933535](https://github.com/firenook-dev/firenook/actions/runs/39450933535)). - A consumer's gates passed on the candidate built from the PR head (`0.2.0-local.ge8ecd3325537`): Extensions admission (24 handlers admitted, 2 ignored upstream, as the official emulator records), the installed package smoke and its 26 integration tests. Its stack with three Extensions was ready in 2.6 s, against 7.4 s on 0.2.0-next.6. - Published straight to `latest`: the CLI's `next` stays on 0.2.0-next.6 until it is moved by hand, and a preview install, which follows `next`, is not told about this release until then. [The release on GitHub](https://github.com/firenook-dev/firenook/releases/tag/npm-v0.2.0-next.7). ## The latest is 0.2.0 The first release for public use. Install it, and read what changed on the way. `npm install --save-dev firenook@0.2.0` [Read the 0.2.0 notes](https://firenook.dev/changelog/0.2.0)