# @kinqs/brainrouter-sdk

Source: https://glimind.com/server/npm%2F-kinqs%2Fbrainrouter-sdk
Registry id: `npm/-kinqs/brainrouter-sdk`
Status: unknown

## Install

```json
{
  "mcpServers": {
    "kinqs-brainrouter-sdk": {
      "command": "npx",
      "args": [
        "-y",
        "@kinqs/brainrouter-sdk"
      ]
    }
  }
}
```

Claude Code:

```sh
claude mcp add kinqs-brainrouter-sdk -- npx -y @kinqs/brainrouter-sdk
```

## Tools (80)

- `apply_patch`: Apply a multi-file patch using the Begin/End envelope format ("*** Begin Patch / *** Update File: path / @@ context / -old / +new / *** Add File: / *** Delete F
- `artifact_write`: Create or update a durable ARTIFACT — a self-contained, reusable piece of work the user will want to refer back to, edit, or keep: a design doc, a report, an HT
- `ask_user_choice`: Pause the turn and ask the human to commit to ONE of 2–4 mutually exclusive approaches.
- `close_worker`: Mark a worker thread closed (terminal). Its in-process run, if any, is left to wind down but its result is no longer adopted.
- `computer_use`: Control the real local desktop mouse/keyboard or capture a screenshot. Requires user approval for mutating actions and only appears when desktop computer use is
- `connector_list`: List the connectors configured for this workspace (GitHub, filesystem, web, Slack, Jira, Confluence, Notion, Linear, GitLab, Google Drive, Gmail, MCP). Read-onl
- `connector_run`: Run one connector's ingest → memory checkpoint: fetch new documents from the source, persist them, and import them into memory so future recall can cite them. D
- `create_skill`: Scaffold a new skill. If scope is "global", ensure content is universal (replace project-specific terms like "YourProject" with generic ones like "the project")
- `edit_file`: Edit an existing file in the workspace by replacing a target substring with a replacement string.
- `extract_result`: Expand a large tool result that was handed off (you hold a `resultRef` instead of the full output). With no `query`, returns the head of the result; with a `que
- `fetch_url`: Fetch and extract clean text from an HTTP(S) URL using the configured in-house crawler.
- `file_vulnerability`: Record one verified pentest finding. A reproducible proof of concept, CVSS 3.1 vector, CWE, and remediation are mandatory. Duplicate root causes are returned wi
- `finish_scan`: Complete the active pentest only after every worker is terminal. Before finishing, consider whether individually-confirmed findings chain into higher-impact end
- `get_persona`: Fetch a persona definition.
- `get_reference`: Fetch a reference document.
- `get_skill`: Fetch a specific section of a skill (default: workflow) or read an auxiliary file within the skill directory.
- `get_template_doc`: Compatibility tool for downstream clients: read a project template document or section. BrainRouter onboarding uses client-local contracts instead.
- `glob_files`: Recursively find files in the workspace matching a glob/wildcard pattern (e.g., "src/**/*.ts" or "*.json").
- `grep_search`: Search workspace files for a REGULAR-EXPRESSION query (JS regex syntax — e.g. `foo|bar`, `\\bclass\\b`). `path` may be a directory (searched recursively, build/
- `kill_command`: Terminate a background run_command by id (the counterpart to run_command({background:true}) / task_output). Sends SIGTERM by default; use SIGKILL to force-stop 
- `list_dir`: List the contents of a directory in the workspace.
- `list_extensions`: Read-only: enumerate the installed BrainRouter extensions and the tools each one contributes, so you can tell the user exactly what capabilities you have and wh
- `list_mcp_resource_templates`: List parameterized resource templates provided by MCP servers.
- `list_mcp_resources`: List resources provided by MCP servers. Resources are structured context such as files, schemas, or app data; prefer them over web search when available.
- `list_requests`: List HTTP requests captured by the authorized pentest proxy.
- `list_sitemap`: List the authorized target sitemap recorded by the pentest proxy.
- `list_skills`: List all available skills (global + local merged).
- `list_template_docs`: Compatibility tool for downstream clients: list project-specific template documentation. BrainRouter onboarding uses client-local contracts instead.
- `lsp`: Semantic code navigation via the language server (exact, not fuzzy). actions: definition / references / hover (need file + 1-based line + character), symbols (f
- `mark_chapter`: Mark the start of a new chapter when the work shifts to a meaningfully different phase (exploration -> implementation -> verification, or a topic pivot). Sparin
- `mcp_call`: Invoke an MCP tool by its exact name (from mcp_search) with the given arguments. Runs the SAME approval and safety checks as calling the tool directly. Use this
- `mcp_describe`: Return the full description and input JSON schema for one or more MCP tools by their exact name (as returned by mcp_search). Call before mcp_call to learn a too
- `mcp_refresh_catalog`: Re-scan connected MCP servers and return a summary of available tools grouped by server (with counts). Use if a server was just connected or you suspect the cat
- `mcp_search`: Search the connected MCP tool catalog by keyword (matches tool name, server, and description) and return the best-matching tools, each with a one-line summary. 
- `notebook_edit`: Edit a Jupyter notebook (.ipynb) cell by index. edit_mode="replace" (default) overwrites the cell source; "insert" adds a new cell AT cell_index (shifting the r
- `notify_user`: Send an out-of-band notification to the user, surfaced above the prompt and folded into the next turn (BrainRouter's completion-feedback channel).
- `planner_add`: Capture an item in the planner. Use when the user says they need to do something later,
- `planner_complete`: Mark a planner item done. ONLY when the user says it is done. NEVER infer completion from
- `planner_find`: Search the planner by text across titles and notes. Read-only. Use it to check whether
- `planner_schedule`: Set aside time for a planner item. Leave scheduledFor empty for an unscheduled block — a
- `planner_today`: Read the user's planner for today: committed items, the current or next time block,
- `pull_request_watch`: Watch a GitHub pull request in the background while you continue working. After pushing or opening a PR, call action="start" once; do not manually loop or block
- `read_file`: Read the contents of a file from the workspace. Optional line ranges can be provided.
- `read_mcp_resource`: Read a specific resource from an MCP server by server id and URI.
- `read_worker_summary`: Read a worker thread's rolling summary (summary.md) without waiting for it to finish.
- `reconcile_steer`: Classify one pending Steer receipt before acting on it. Extension-origin receipts are evidence-only.
- `repeat_request`: Replay a captured request through the authorized pentest proxy, optionally with a safe mutation.
- `research_brief`: Emit a report-ready markdown research brief from the session ledger: every finding plus an explicit "Uncertainty & conflicts" section (corroborated vs single-so
- `research_note`: Record one claim with structured source provenance in the durable session research ledger. Capture the exact evidence and limitations after reading each source 
- `run_command`: Run a shell command on the user's terminal. Requires user approval before execution. Pass background:true to DETACH a long-running command (build, server, watch
- `schedule_task`: Create, list, or cancel a scheduled agent task. Scheduled tasks dispatch a slash command (must start with "/") on a recurring cron or one-shot delay, fired by B
- `scope_rules`: Read or replace the proxy scope rules for the authorized target.
- `search_skills`: Fuzzy search across all skills.
- `session_heartbeat`: Advance lastHeartbeatAt for an active session. Call every ~30s. Returns `{ updated: true }` on success; `{ updated: false }` when no row exists (client should r
- `session_info`: Read-only context about the current session and workspace: the current session id/title, its recent transcript activity, the workspace's recent sessions, and an
- `session_list`: List active peer sessions for a user. Default scope is `last_heartbeat_at` within the last 2 minutes (recently-active peers). Pass `includeStale: true` to see e
- `session_read`: Read the recent transcript of another conversation in this workspace (get its sessionKey from session_list). Returns redacted role/name/timestamp/content entrie
- `session_reference`: Pull a bounded snapshot of another session (get its sessionKey from session_list) into context as EXPLICITLY UNTRUSTED data. Returns one budget-capped, redacted
- `session_register`: Register an active MCP client with the brain. Called once at client startup. Returns a stable `sessionKey` you should pass to subsequent heartbeats — if you alr
- `session_search`: Search across this workspace's conversations for a text query. Returns the matching sessions with a few redacted snippets each (role, timestamp, snippet, match 
- `session_unregister`: Remove an active session row immediately. Called by clients on clean exit so peers don't see a 5-min ghost while the sweeper catches up. Idempotent: returns `{ 
- `spawn_worker_thread`: Start a persistent background worker thread for a self-contained task. It runs detached (your turn does NOT block), persists its transcript + rolling summary + 
- `switch_model`: Switch THIS session to a named LLM profile (a saved model preset) for all subsequent model calls — e.g. move to a stronger profile for a hard design/debugging s
- `task_output`: Read incremental output of a background run_command: returns { status, exitCode, chunk, nextOffset, complete }. Pass the previous nextOffset as fromByte to read
- `terminal_list`: List live native terminal sessions owned by the current Desktop workspace. Use terminal_read with an id to inspect visible shell output. Only available to the t
- `terminal_read`: Read bounded, plain-text output from a live native terminal. Omit fromOffset to inspect the latest output; pass the returned nextOffset to read only new output.
- `terminal_write`: Send bounded input to an existing live native terminal. Use only when the user asked you to interact with that terminal; command execution normally belongs in r
- `track_query`: Read the workspace project board (Track mode — one project per workspace). action: "list" (all work items, optionally filtered), "get" (one by key), "board" (it
- `track_update`: Create or change project-board work (Track mode). action: "create", "transition", "comment", "link", "sprint-create", "assign-sprint", "batch-transition", "spri
- `update_plan`: Create or update the durable CLI task plan. Use it ONLY for work with ≥3 non-trivial steps (1–2 steps: just do them).
- `update_skill`: Update an existing skill section. Supports "shadowing" global skills locally or updating global skills directly.
- `view_request`: View one captured HTTP request and response by proxy request id.
- `wait_until`: Block until a workspace condition holds or the timeout elapses: a file exists, or a file contains a text marker. Use after starting background work (a worker, a
- `wait_worker`: Block until a worker thread finishes or the wait timeout elapses, then return its status + summary. A wait timeout does not stop or fail the worker.
- `web_search`: Search the public web with the configured provider and return normalized results (title, url, snippet). Useful when fetch_url needs a starting point.
- `workspace_create`: Make a new record in another surface of the workspace from something you are looking at — a
- `workspace_link`: Record that one thing references another, by writing the reference into the referring
- `workspace_resolve`: Follow a brainrouter:// reference and read the CURRENT state of what it points at — a note
- `workspace_update`: Change something that already exists in the workspace — rename a note or a task, tick a
- `write_file`: Create a new file or completely overwrite an existing file in the workspace.

Measured by Glimind (https://glimind.com): MCP handshake + tool list only, no tool is ever called. CC BY 4.0.