Skip to main content
Glama

nsnswewe

N S N S W E W E: a compass needle swinging before it settles. Install nsnswewe, then type strategy.

Your strategy as data — four things, and only four. See it, look at one position and its case, decide on it, add one. The same four on a small web page, on the command line and over MCP.

MIT licensed. Runs entirely on your machine, against SQLite or Postgres.

Why so small

Teams decide things constantly and then lose them: the decision lives in a task title, a closed issue, a doc from March, and six weeks later it is re-argued from scratch. A strategy you can see on one page is one you can check work against. Simple is explainable; this does the explainable part.

Related MCP server: Hegelion

The four things

The strategy is laid out in the five boxes of Playing to Win (Lafley & Martin): winning aspiration, where to play, how to win, capabilities, management systems. A position is one claim in one box. It is undecided until you adopt it (I hold this) or reject it (I do not, and here is why); either is dated. What backs it is its evidence: notes of what was actually observed, each saying which way it cuts — supports, opposes, or complicates.

Web

CLI

MCP tool

1. See the strategy

/

strategy list

list

2. One position and its evidence

/positions/<id>

strategy show <id>

show

3. Decide: adopt · reject · note

forms on the position page

strategy adopt <id> · strategy reject <id> --note … · strategy evidence add <id> --summary … [--stance supports]

adopt · reject · evidence_add

4. Add a position into a box

/add

strategy add --title … --box "where to play"

add

Several strategies in one store. A cascade is the five boxes for one project, or for one pass at the same question. Create, rename and delete them on the Cascades page, with strategy cascade create|list|show|rename|delete, or with the MCP tools cascade_list, cascade_create, cascade_rename and cascade_delete. The strategy page has a dropdown to show one (and, when some positions are in no cascade, "Unassigned positions", also --cascade -); strategy list --cascade KEY and the list tool's cascade do the same, and add into a cascade puts the new position in it. Deleting a cascade never deletes a position.

The CLI also does the housekeeping: strategy init, strategy export, strategy import, and strategy web to serve the pages.

Every surface reads through queries.py and writes through core.py, so they cannot disagree: adopting dates the position, a rejection must carry a reason, a position must name one of the five boxes.

Quick start

uv tool install 'nsnswewe[web,mcp]'   # → strategy, strategy-mcp (and nsnswewe, the same as strategy)
strategy init                  # SQLite at ~/.strategy/store.db, or set STRATEGY_DB_URL
strategy add --title "Sell to teams, not individuals" --box "where to play"
strategy list
strategy web                   # http://127.0.0.1:8021

To try it without installing: uvx nsnswewe init, then uvx nsnswewe list.

From a clone of this repo, uv tool install '.[web,mcp]' does the same as the install line.

For a chat client, run strategy-mcp (stdio). It has the four things and no housekeeping.

Configuration

Variable

Purpose

Default

STRATEGY_DB_URL

Where the store lives

SQLite at ~/.strategy/store.db

STRATEGY_EXPORT_PATH

JSONL backup target

~/.strategy/positions.jsonl

Settings are read from the process environment first, then ~/.env, then the default. ${PART}-style references in those files resolve the way a shell would; a reference nothing defines drops the key with a line on stderr rather than leaving half a connection string.

If STRATEGY_DB_URL is set and cannot be opened, the CLI fails and says so. It never quietly falls back to the local SQLite copy.

Your positions are not in this repository. They name real commitments and real people, and live in your store.

The web page

Four pages (the strategy, a position, add, cascades) and /health, bound to loopback with no login. Writes are form posts that go through core; a cross-origin post is refused.

Publishing hygiene

scripts/check_publishable.py fails on real Vikunja task ids, patent numbers, or terms from a private list kept outside the repo (STRATEGY_PRIVATE_TERMS; plain substrings, or re: lines for patterns). tests/test_publishable.py runs the same check in the suite.

License

MIT — see LICENSE. Pinned to Python 3.12 by .python-version.

Available Tools

10 tools
addB

Add a position into a box — 1 aspiration, 2 where to play, 3 how to win, 4 capabilities, 5 management systems (number or name). It arrives undecided. Optional cascade: add it to that strategy.

Inputs: box (required), title (required), source, cascade, +1 more. Returns: object

adoptC

Adopt a position: I hold this. Dates it today.

Inputs: position_id (required), note. Returns: object

cascade_createB

Start a new strategy (cascade) by name, with the question it answers (optional). It starts empty.

Inputs: name (required), question. Returns: object

cascade_deleteA

Delete a cascade. Its positions are kept, with their decisions and evidence, as unassigned positions (list with cascade "-").

Inputs: cascade (required). Returns: object

cascade_listA

The strategies (cascades) in this store: key, name, the question each answers, and how many positions it holds. Pass a key as cascade to list or add.

Inputs: no named parameters. Returns: result

cascade_renameB

Rename a cascade, or change the question it answers. Its key stays the same.

Inputs: cascade (required), name, question. Returns: object

evidence_addC

Note something observed that bears on a position; optional stance: supports | opposes | complicates.

Inputs: summary (required), position_id (required), source, stance. Returns: object

listA

The strategy: every current position under its Playing to Win box, aspiration first; optional cascade (a key from cascade_list) for one strategy, or "-" for the positions no cascade holds. To sort, triage or prioritise tasks in any tracker: call list first and judge each task against the ADOPTED positions only; if none is adopted, stop and say so. Keep what serves one, park what doesn't, and name the position that decided it. Ask before changing anything in the tracker, and never delete a task.

Inputs: cascade. Returns: result

rejectC

Reject a position: I do not hold this. A reason is required.

Inputs: note (required), position_id (required). Returns: object

showC

One position: its box, whether it is adopted, and the evidence noted on it.

Inputs: position_id (required). Returns: object

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.1/5.0

    Scored across 10 tools

    Disambiguation4/5

    Most tools target clearly different operations: list/show for reading positions, add for creation, adopt/reject for decisions, evidence_add for evidence, and cascade_* for strategy containers. The only mild overlap is between list and cascade_list, and between list and show, but the descriptions clarify resource and scope well enough.

    Naming Consistency3/5

    Core position operations are bare verbs (list, show, add, adopt, reject), while cascade operations use a noun_verb prefix and evidence_add uses a different noun_verb ordering. It remains readable, but there is no single predictable naming pattern across the whole set.

    Tool Count5/5

    Ten tools is well within a healthy range for this domain, and each tool maps to a distinct operation in the strategy/cascade model. None feels redundant or purely decorative.

    Completeness3/5

    The surface covers position creation/reading, decisions, evidence, and full cascade CRUD. However, there is no way to update or delete a position (for example, editing its text or moving it between Playing to Win boxes), which is a notable lifecycle gap.

    Maintenance

    ActivitySlowing
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Provides 8 MCP tools for deterministic, read-only reasoning: intake, routing, planning, rubric, sweep checklist, verdict gate, reflection, and evaluation. It forces scope locks, disconfirmation-first plans, blind-spot sweeps, and evidence-gated verdicts.
      Apache 2.0
    • F
      license
      Not graded
      quality
      A
      maintenance
      Enables MCP-enabled editors and agents to apply dialectical reasoning and player-coach autocoding workflows, producing structured thesis–antithesis–synthesis analyses and independently verified code implementations. It provides tools for initiating sessions, executing turns, and persisting or restoring reasoning and coding state.
      -
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables MCP clients to search, record, correct, and review the provenance and version history of memories, and to obtain answers composed from guarded context in which private values are replaced by placeholders and resolved locally on the device. Exposes these capabilities through dependency-free tools over stdio.
      MIT