Update dev and user-facing docs for tool consolidation

- TESTING.md: move_cells → copy_cells(delete_source=true), get_kernel_variables → kernel_variables
- COMPARISON.md: reflect kernel management tools, remove outdated "more kernel tools" advantage
- CHANGELOG.md: add [Unreleased] section documenting 55→41 consolidation
- docs changelog: same unreleased section for user-facing docs

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: ianhi

CHANGELOG.md

@@ -2,6 +2,32 @@
 
 All notable changes to the jupyterlab-collab-mcp.
 
+## [Unreleased]
+
+### Changed
+- **Consolidated 55 tools down to 41** — reduced schema token overhead by merging related tools:
+  - `insert_and_execute`/`update_and_execute` → `insert_cell(execute=true)`/`update_cell(execute=true)`
+  - `delete_cells` → `delete_cell` (now accepts `indices`, `cell_ids`, `start_index`/`end_index`)
+  - `move_cells` → `copy_cells(delete_source=true)`
+  - `get_cell_metadata`/`set_cell_metadata` → `cell_metadata` (omit `metadata` to GET, provide to SET)
+  - `get_notebook_metadata`/`set_notebook_metadata` → `notebook_metadata` (same pattern)
+  - `add_cell_tags`/`remove_cell_tags` → `cell_tags(action="add"/"remove")`
+  - `lock_cells`/`unlock_cells`/`list_locks` → `cell_locks(action="acquire"/"release"/"list")`
+  - `snapshot_notebook`/`restore_snapshot`/`list_snapshots`/`diff_snapshot` → `snapshot(action="save"/"restore"/"list"/"diff")`
+  - `get_kernel_status`/`interrupt_kernel`/`restart_kernel` → `kernel(action="status"/"interrupt"/"restart")`
+  - `get_kernel_variables`/`inspect_variable` → `kernel_variables` (provide `names` to inspect)
+- Removed `insertCell` parameter from `execute_code` — use `insert_cell(execute=true)` instead
+- Removed `max_output_lines`, `output_tail`, `output_grep` from execute tools — use `filter_output` instead
+
+### Added
+- `filter_output` tool — post-process cached execution results with grep, head, tail, max_lines
+- `show_diff` parameter on `update_cell` for inline diff display
+- Execution result caching for `filter_output` access to full unfiltered output
+
+### Fixed
+- Batch delete (`delete_cell` with indices/range) now records change history for `recover_cell`
+- Cross-notebook `copy_cells` response clarifies that returned IDs are new destination cell IDs
+
 ## [0.8.0] - 2025-02-11
 
 ### Added

COMPARISON.md

@@ -15,7 +15,7 @@ This document compares our implementation with the existing [datalayer/jupyter-m
 | Multi-notebook | Implicit (by path) | Explicit `use_notebook`/`unuse_notebook` |
 | File browser | `list_files` | `list_files` |
 | Open/create notebooks | `open_notebook`, `create_notebook` | Similar |
-| Kernel management | No | `list_kernels`, `restart_notebook` |
+| Kernel management | `kernel` (status/interrupt/restart), `kernel_variables` | `list_kernels`, `restart_notebook` |
 | Output filtering | `output_format`, `cell_type` | `response_format` (brief/detailed) |
 | Streaming execution | No | Yes (`stream` flag) |
 | JupyterLab commands | No | `run-all-cells`, `get-selected-cell` |
@@ -86,11 +86,7 @@ They have `response_format` (brief/detailed) but less granular control.
 - Better for team/production deployments
 - Survives client disconnects
 
-### 2. More Kernel Tools
-- `list_kernels` - See available kernels
-- `restart_notebook` - Restart kernel without reconnecting
-
-### 3. Streaming Execution
+### 2. Streaming Execution
 ```python
 execute_cell(cell_index, stream=True, progress_interval=1.0)
 ```

TESTING.md

@@ -3,7 +3,7 @@
 When testing multi-agent collaboration on notebooks, use a team of 4+ agents working simultaneously. The test should exercise:
 
 1. **Cell ID stability**: Agents use `cell_id` (not indices) for all operations. One agent inserting cells mid-notebook must not break another agent's references.
-2. **Cross-notebook operations**: Agents work across multiple notebooks using `copy_cells` and `move_cells`. E.g., one agent builds a data pipeline notebook while another builds a visualization notebook, and they share cells between them.
+2. **Cross-notebook operations**: Agents work across multiple notebooks using `copy_cells` (and `copy_cells` with `delete_source=true` for moves). E.g., one agent builds a data pipeline notebook while another builds a visualization notebook, and they share cells between them.
 3. **Execute range**: Agents use `execute_range` to run multi-cell sections, not just single cells.
 4. **Multi-plot cells**: Include cells that produce multiple matplotlib figures (subplots, figure galleries) to stress-test `max_images`/`include_images` context management. Agents should use `max_images=2` or `include_images=false` for plot-heavy cells to conserve context.
 5. **Concurrent inserts**: Multiple agents inserting cells in the same notebook simultaneously — cell IDs prevent index collisions.
@@ -16,7 +16,7 @@ The team lead acts as a **scientist/observer**, not a manager:
 - **Observe**: Monitor the change log and notebook state for high-level issues
 - **Don't micromanage**: Let agents coordinate organically through messaging and kernel variable polling. Don't direct traffic, assign work to idle agents, or prevent conflicts — conflicts are valuable for stress testing
 - **Intervene only for systemic issues**: e.g., a shared kernel crash, a tool bug blocking all agents, or a fundamental misunderstanding of the task
-- **Avoid rigid dependency graphs**: Use `blockedBy` sparingly or not at all. Let agents discover data availability themselves via `get_kernel_variables` and communicate through messages. Organic coordination surfaces real collaboration pain points that artificial sequencing hides.
+- **Avoid rigid dependency graphs**: Use `blockedBy` sparingly or not at all. Let agents discover data availability themselves via `kernel_variables` and communicate through messages. Organic coordination surfaces real collaboration pain points that artificial sequencing hides.
 
 ## Test Phases
 

docs/src/content/docs/changelog.md

@@ -5,6 +5,29 @@ description: Version history for jupyterlab-collab-mcp.
 
 All notable changes to the jupyterlab-collab-mcp.
 
+## [Unreleased]
+
+### Changed
+- **Consolidated 55 tools down to 41** — reduced schema token overhead by merging related tools:
+  - `insert_and_execute`/`update_and_execute` → `insert_cell(execute=true)`/`update_cell(execute=true)`
+  - `delete_cells` → `delete_cell` (now accepts `indices`, `cell_ids`, `start_index`/`end_index`)
+  - `move_cells` → `copy_cells(delete_source=true)`
+  - `get/set_cell_metadata` → `cell_metadata`, `get/set_notebook_metadata` → `notebook_metadata`
+  - `add/remove_cell_tags` → `cell_tags(action="add"/"remove")`
+  - `lock/unlock/list_locks` → `cell_locks(action="acquire"/"release"/"list")`
+  - 4 snapshot tools → `snapshot(action="save"/"restore"/"list"/"diff")`
+  - 3 kernel tools → `kernel(action="status"/"interrupt"/"restart")`
+  - `get_kernel_variables`/`inspect_variable` → `kernel_variables`
+
+### Added
+- `filter_output` tool — post-process cached execution results with grep, head, tail
+- `show_diff` parameter on `update_cell`
+- Execution result caching
+
+### Fixed
+- Batch delete now records change history for `recover_cell`
+- Cross-notebook `copy_cells` clarifies destination cell IDs
+
 ## [0.8.0] - 2025-02-11
 
 ### Added