ติดตั้ง Bantamkit อย่างไร?

คัดลอกคอนฟิก stdio สำหรับ Cursor, Claude Desktop, Claude Code หรือ VS Code ได้ในหน้านี้ คอนฟิกนี้ใช้การติดตั้งของผู้เผยแพร่ ไม่ใช่พร็อกซีของ PlugMCP

Bantamkit ต้องใช้ API key ไหม?

ระเบียนนี้ไม่ได้ระบุตัวแปรสภาพแวดล้อมที่จำเป็น

Bantamkit ใช้ทรานสปอร์ตแบบใด?

Bantamkit ทำงานแบบโลคัลผ่าน stdio

Bantamkit ถูกตรวจสอบล่าสุดเมื่อใด?

ตรวจสอบล่าสุดเมื่อ 2026-11-14 Persistent memory for coding agents: an MCP server for Claude Code, Claude Desktop, Cursor and VS Code Copilot. Layered project memory, Claude Code hooks, JSON Schema validation, token ledger and shift-work tools. Pure Node; install once with npx, run off

เซิร์ฟเวอร์ MCP Bantamkit

Power your agent

เพิ่มเซิร์ฟเวอร์นี้

คำสั่งของผู้เผยแพร่เอง ไม่ใช่เอนด์พอยต์ของ PlugMCP

เพิ่มใน Cursor เพิ่มใน VS Code รันบน PlugMCP

Host a copy with our keys and logs.

คำสั่งแบบโลคัล

npx -y [email protected]

Claude Code ไม่มีลิงก์ติดตั้งสำหรับคำสั่งแบบโลคัล ให้คัดลอกคำสั่งด้านล่างแล้วรันในเทอร์มินัล

shClaude Code

1claude mcp add bantamkit-mcp -- npx -y bantamkit-mcp@0.39.3

รันในเทอร์มินัล

jsonClaude Desktop

1{2  "mcpServers": {3    "bantamkit-mcp": {4      "command": "npx",5      "args": [6        "-y",7        "[email protected]"8      ]9    }10  }11}

claude_desktop_config.json

โลคัล · npmการติดตั้งจากผู้เผยแพร่ทีละขั้นตอน: เพิ่มใน Claude

URL แบบรีโมตหรือคำสั่งบนเครื่องของคุณ? การติดตั้งสองแบบต่างกันอย่างไร

รันบน PlugMCP

Host a copy with our keys and logs.

A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.

เปิด

ข้อมูลเซิร์ฟเวอร์

โลคัล

ทรานสปอร์ต: stdio
เวอร์ชัน: 0.39.3
ตรวจสอบล่าสุด

จากผู้เผยแพร่

เอกสารประกอบ

bantamkit — memory MCP server for Claude Code, Cursor, VS Code Copilot and Claude Desktop

Topic	What you'll find
Install the MCP server	Three ways to set it up — pick one
Install once, run offline (recommended)	One command; later launches need no network
Run with npx at every launch	The online form, and why it hangs offline
Install with Python (PyPI)	A venv you keep; a wheelhouse for no-network machines
Connect to a host	The same five steps for every host
Connect to Claude Code	`--install claude`, `claude mcp list`, `claude mcp remove`
Connect to Claude Desktop	`claude_desktop_config.json` per OS
Connect to Cursor	`~/.cursor/mcp.json`
Connect to VS Code (GitHub Copilot)	`mcp.json` with the `servers` key
Connect other MCP clients	Codex and any stdio host
Configuration	Every flag and environment variable, with its default
Update	`--update`, then restart the server
Troubleshooting	Timeouts, Connection closed, refusals, the wrong store
What it serves	15 tools, three prompts, two resource templates
Use it as a Python library	`pip install bantamkit` and a minimal agent
Recommended defaults	Which components to attach, measured on four models
Documentation	Every page under `docs/`
Repo layout	What lives where

Route	`command`	`args`
npm, install once	`/absolute/path/to/node`	`["/Users/you/.bantamkit/mcp/node_modules/bantamkit-mcp/dist/cli.js"]`
Python venv	`/absolute/path/to/env/bin/bantamkit-mcp`	`[]`
npx every launch	`npx`	`["-y", "bantamkit-mcp"]`

Setting	What it does	Default	Example
`BANTAMKIT_MEMORY_DIR`	Pins the store to one absolute path; a missing or relative path refuses at startup	unset: nearest existing `.bantamkit/memory` at or above the start directory	`"env": {"BANTAMKIT_MEMORY_DIR": "/abs/project/.bantamkit/memory"}`
`--store STORE`	One store, layering off; outranks `BANTAMKIT_MEMORY_DIR`	off (layered)	`"--store", "/abs/store"`
`--start START`	Where store discovery starts; not with `--store`	cwd	`"--start", "/abs/project"`
Layered memory	Recall reads the project store, stores granted in `.bantamkit/config.yaml`, and `~/.bantamkit/memory`; saves go to the project store	on	docs/memory.md → Layers
`--k K`	Default recall budget	`3`	`"--k", "5"`
`--index-budget BYTES`	Memory index byte budget	`24000`	`"--index-budget", "32000"`
`BANTAMKIT_EVENT_LOG`	Logs tool outcomes as JSONL	off; `on` → `<store>/events/mcp.jsonl`; other values are a path	`"env": {"BANTAMKIT_EVENT_LOG": "on"}`
`BANTAMKIT_RETENTION`	The daily Retention job: deletes session read ledgers older than 7 days, folds log lines older than 30 days, archives closed `.shiftwork` jobs 7 days after they close (verified before delete). It does not stop the daily skill audit, which has its own switch (next row)	on; `off` (or `0`/`false`/`no`) turns it off	`"env": {"BANTAMKIT_RETENTION": "off"}` — docs/storage.md
`BANTAMKIT_SKILL_AUDIT`	The daily skill audit (#205): the same once-a-day child audits your skills and writes `~/.bantamkit/skill-audit.json`; `bantamkit_status` prints its last run. Its own switch: with `BANTAMKIT_RETENTION=off` and this on, the child still wakes once a day, audits, and deletes nothing. Off also silences the SessionStart skill-audit line, even over an older record; the `skill_audit` prompt still runs on request	on; `off` (or `0`/`false`/`no`) turns it off	`"env": {"BANTAMKIT_SKILL_AUDIT": "off"}` — docs/storage.md
`BANTAMKIT_SKILL_AUTO`	Skills from experience (#206): lets the daily run INSTALL a skill it mined (twice the bar, a gated draft, one a day, five live, 30 days of probation) and make the low-risk EDITS to installed skills (keyword widening, a Pitfalls line, a drafted description) and PROMOTE a skill used 3 times in 30 days to a native link, on its own. Off, the same work waits for you as proposals in the `skill_propose` prompt; the judge still rolls back a change of bantamkit's that made things worse. `bantamkit_status`'s `skills:` line ends with `auto on` or `auto off (BANTAMKIT_SKILL_AUTO=on to opt in)`	off in 0.37.0; `on` (or `1`/`true`/`yes`, any case) opts in, and any other value leaves it off	`"env": {"BANTAMKIT_SKILL_AUTO": "on"}`
`BANTAMKIT_LLM_URL`, `BANTAMKIT_LLM_MODEL`	Skills from experience (#206): an OpenAI-compatible endpoint (`<url>/chat/completions`) bantamkit itself may ask to draft a skill, after MCP sampling and before the host agent. Both must be set. No key is sent but `Bearer none`, which local endpoints (Ollama, LM Studio) accept. The daily run drafts through it only when `BANTAMKIT_SKILL_AUTO` is on	unset: no endpoint	`"env": {"BANTAMKIT_LLM_URL": "http://localhost:11434/v1", "BANTAMKIT_LLM_MODEL": "qwen2.5-coder"}` — docs/skills.md
`BANTAMKIT_SKILL_BUDGET`	Skills from experience (#206): the catalogue budget in bytes (the summed `description` bytes of every skill the host loads) that promoting a skill to a native link may not exceed	`8192`; any value that is not 1–15 digits falls back to it	`"env": {"BANTAMKIT_SKILL_BUDGET": "12000"}`
`BANTAMKIT_INDEX_DIR`	Where the derived recall indexes live (hot and cold); deleting them loses nothing	`~/.bantamkit/index`	`"env": {"BANTAMKIT_INDEX_DIR": "/abs/scratch/index"}`
`BANTAMKIT_ASSETS`	Your own asset pack (skills, rubrics, schemas)	the pack inside the package	`"env": {"BANTAMKIT_ASSETS": "/abs/my-assets"}`
`BANTAMKIT_HOST_LOG_ROOT`	Where `--mcp-report` finds the host's MCP logs	macOS `~/Library/Caches/claude-cli-nodejs`; elsewhere unset	`BANTAMKIT_HOST_LOG_ROOT=/abs/logs bantamkit-mcp --mcp-report`
`BANTAMKIT_PRICES`	Price table for the token ledger	`pricing/default.json` in the asset pack	`"env": {"BANTAMKIT_PRICES": "/abs/prices.json"}`
`--which`	Where a checkout launcher resolved its halves — only `tools/bantamkit-mcp` and `tools/bantamkit-mcp-node`, not the npm/PyPI package	—	`tools/bantamkit-mcp --which`

Setting

What it does

Default

Example

BANTAMKIT_MEMORY_DIR

Pins the store to one absolute path; a missing or relative path refuses at startup

unset: nearest existing .bantamkit/memory at or above the start directory

"env": {"BANTAMKIT_MEMORY_DIR": "/abs/project/.bantamkit/memory"}

--store STORE

One store, layering off; outranks BANTAMKIT_MEMORY_DIR

off (layered)

"--store", "/abs/store"

--start START

Where store discovery starts; not with --store

cwd

"--start", "/abs/project"

Layered memory

Recall reads the project store, stores granted in .bantamkit/config.yaml, and ~/.bantamkit/memory; saves go to the project store

docs/memory.md → Layers

--k K

Default recall budget

3

"--k", "5"

--index-budget BYTES

Memory index byte budget

24000

"--index-budget", "32000"

BANTAMKIT_EVENT_LOG

Logs tool outcomes as JSONL

off; on → <store>/events/mcp.jsonl; other values are a path

"env": {"BANTAMKIT_EVENT_LOG": "on"}

BANTAMKIT_RETENTION

The daily Retention job: deletes session read ledgers older than 7 days, folds log lines older than 30 days, archives closed .shiftwork jobs 7 days after they close (verified before delete). It does not stop the daily skill audit, which has its own switch (next row)

on; off (or 0/false/no) turns it off

"env": {"BANTAMKIT_RETENTION": "off"} — docs/storage.md

BANTAMKIT_SKILL_AUDIT

The daily skill audit (#205): the same once-a-day child audits your skills and writes ~/.bantamkit/skill-audit.json; bantamkit_status prints its last run. Its own switch: with BANTAMKIT_RETENTION=off and this on, the child still wakes once a day, audits, and deletes nothing. Off also silences the SessionStart skill-audit line, even over an older record; the skill_audit prompt still runs on request

on; off (or 0/false/no) turns it off

"env": {"BANTAMKIT_SKILL_AUDIT": "off"} — docs/storage.md

BANTAMKIT_SKILL_AUTO

Skills from experience (#206): lets the daily run INSTALL a skill it mined (twice the bar, a gated draft, one a day, five live, 30 days of probation) and make the low-risk EDITS to installed skills (keyword widening, a Pitfalls line, a drafted description) and PROMOTE a skill used 3 times in 30 days to a native link, on its own. Off, the same work waits for you as proposals in the skill_propose prompt; the judge still rolls back a change of bantamkit's that made things worse. bantamkit_status's skills: line ends with auto on or auto off (BANTAMKIT_SKILL_AUTO=on to opt in)

off in 0.37.0; on (or 1/true/yes, any case) opts in, and any other value leaves it off

"env": {"BANTAMKIT_SKILL_AUTO": "on"}

BANTAMKIT_LLM_URL, BANTAMKIT_LLM_MODEL

Skills from experience (#206): an OpenAI-compatible endpoint (<url>/chat/completions) bantamkit itself may ask to draft a skill, after MCP sampling and before the host agent. Both must be set. No key is sent but Bearer none, which local endpoints (Ollama, LM Studio) accept. The daily run drafts through it only when BANTAMKIT_SKILL_AUTO is on

unset: no endpoint

"env": {"BANTAMKIT_LLM_URL": "http://localhost:11434/v1", "BANTAMKIT_LLM_MODEL": "qwen2.5-coder"} — docs/skills.md

BANTAMKIT_SKILL_BUDGET

Skills from experience (#206): the catalogue budget in bytes (the summed description bytes of every skill the host loads) that promoting a skill to a native link may not exceed

8192; any value that is not 1–15 digits falls back to it

"env": {"BANTAMKIT_SKILL_BUDGET": "12000"}

BANTAMKIT_INDEX_DIR

Where the derived recall indexes live (hot and cold); deleting them loses nothing

~/.bantamkit/index

"env": {"BANTAMKIT_INDEX_DIR": "/abs/scratch/index"}

BANTAMKIT_ASSETS

Your own asset pack (skills, rubrics, schemas)

the pack inside the package

"env": {"BANTAMKIT_ASSETS": "/abs/my-assets"}

BANTAMKIT_HOST_LOG_ROOT

Where --mcp-report finds the host's MCP logs

macOS ~/Library/Caches/claude-cli-nodejs; elsewhere unset

BANTAMKIT_HOST_LOG_ROOT=/abs/logs bantamkit-mcp --mcp-report

BANTAMKIT_PRICES

Price table for the token ledger

pricing/default.json in the asset pack

"env": {"BANTAMKIT_PRICES": "/abs/prices.json"}

--which

Where a checkout launcher resolved its halves — only tools/bantamkit-mcp and tools/bantamkit-mcp-node, not the npm/PyPI package

—

tools/bantamkit-mcp --which

Symptom	Fix
The host times out; the server never answers	The entry uses `npx` with a registry spec and the network is down. Use Install once, run offline
`ENOENT` from the host	A GUI host does not read your shell rc, so `npx`/`node` is not on its PATH. `--install` records absolute paths
Connection closed at startup	Older builds crashed when a GUI host started them in `/`. Update; see docs/mcp.md
`already has a bantamkit entry with different settings … re-run with --force to replace it`	Re-run with `--force`; the old file is kept as `.backup-<date>` (`.backup-<date>.1` and on for a later write the same day)
`MCP server bantamkit already exists in user config`	`claude mcp remove bantamkit -s user`, then install again
The server stopped launching after a node change	`npx -y bantamkit-mcp@latest --install <host> --force` with the node you have now
Recall finds nothing, or the wrong store	Set `BANTAMKIT_MEMORY_DIR`; the reply names the store it searched (the three states)

Tool	What it does
`memory_save`	store one durable fact, deduped and budgeted
`memory_recall`	retrieve facts matching a query
`memory_compact`	archive the stalest facts to fit the index budget
`memory_dream`	consolidate facts the project and profile layers hold under the same name
`memory_queries`	list the facts that have no likely-prompt queries yet, and store the queries written for them; recall matches a prompt against them
`validate_json`	validate a document against a JSON Schema
`skill_audit`	audit a skill catalogue for findings
`shiftwork_clock_in`	open a unit of work and get its brief
`shiftwork_clock_out`	close a unit with status and accounting
`shiftwork_status`	read-only progress summary of a checkpoint
`shiftwork_plan`	read-only: which units of a checkpoint its `depends_on` graph permits to run at once. Never moves the cursor
`work_plan`	turn any `{id, depends_on, priority}` graph into the batches that may run in parallel, plus the widest fan-out
`token_ledger`	what a session cost, read off the host's transcripts
`bantamkit_status`	is bantamkit working, and which build: version, `build_id`, store health against its budget
`build_identity`	report the fingerprint of the source on disk, not the executing code — useful when a machine carries two installs under one name

Page	Covers
Install	Requirements, editable install, endpoints, `BANTAMKIT_ASSETS`, measured MCP install detail
Architecture	The 5-layer model and the no-mixing rule
Usage	Client, agent, tools, components, `structured()`, errors, recommended defaults
Memory	Store layout, the four ops, dedupe and budget, layers, the operator CLI (`python -m bantamkit.memory` / `bantamkit-memory`)
File-access graph	The read ledger and the `file_graph` query tool
Eval	Running the suite, the config matrix, current results
MCP	`bantamkit-mcp` for Claude Code, Codex and any MCP client
Event log	The JSONL outcome log, its vocabulary and the 1 MiB cap
MCP report	`--mcp-report`: the host's MCP log joined with the event log
Status line	`bantamkit-mcp --statusline` and the Claude Code `statusLine` adapter
Hooks	`--install-hooks`: what each Claude Code hook entry does and what it costs
Status	`bantamkit_status`, line by line, and the degraded footer
Token ledger	`token_ledger`: what a session cost, and the price table
Skill audit	`skill_audit`: what the skill catalogue costs every session
Skills from experience	`skill_propose`: skills bantamkit mines from repeated, successful work, the `[skill]` layer, promotion and self-improvement
Shift-work	Checkpoint contract and clock-in/clock-out driver (experimental)
Porting	The Node port, `PyScalar`, and where the two runtimes deliberately differ
Conformance	The two-runtime diff harness and what a `ruling:` pins
Releasing to npm	What ships and the pre-publish checklist

Path	What
`runtime-py/`	The Python runtime (`bantamkit` package) and its test suite
`runtime-ts/`	The pure-Node MCP server (`bantamkit-mcp` on npm)
`assets/`	Language-agnostic asset pack: skills, rubrics, tool schemas, eval tasks
`examples/`	Runnable starter scripts (quickstart, structured output, layered memory)
`tools/`	Repo tools that ship outside the wheel (e.g. the shift-work driver)
`docs/`	This runbook

เพิ่มเซิร์ฟเวอร์นี้

Host a copy with our keys and logs.

ข้อมูลเซิร์ฟเวอร์

เอกสารประกอบ

bantamkit — memory MCP server for Claude Code, Cursor, VS Code Copilot and Claude Desktop

🚀 Quick start

✨ What you get

📑 Contents

📥 Install the MCP server

Install once, run offline (recommended)

Run with npx at every launch

Install with Python (PyPI)

🔌 Connect to a host

Connect to Claude Code

Connect to Claude Desktop

Connect to Cursor

Connect to VS Code (GitHub Copilot)

Connect other MCP clients

🔧 Configuration

🔄 Update

🩺 Troubleshooting

🧰 What it serves

🐍 Use it as a Python library

📊 Recommended defaults

📚 Documentation

📁 Repo layout

Bantamkit

เกี่ยวกับ Bantamkit

เครื่องมือ

คำถาม

ติดตั้ง Bantamkit อย่างไร?

Bantamkit ต้องใช้ API key ไหม?

Bantamkit ใช้ทรานสปอร์ตแบบใด?

Bantamkit ถูกตรวจสอบล่าสุดเมื่อใด?