tm1-mcp-server
[!IMPORTANT] This is a personal testing fork, not the project. The author of tm1-mcp-server is flameY3T1. The project to use, cite, link to and report issues against is flameY3T1/tm1-mcp-server, published on npm as .
接入你的智能体
發布者自己的指令。不是 PlugMCP 端點。
用我們的金鑰與日誌託管一份副本。
npx -y @ffschrattenecker/[email protected]Claude Code 沒有本機指令的安裝連結。複製下方指令並在終端機中執行。
claude mcp add ffschrattenecker-tm1-mcp-server -- npx -y @ffschrattenecker/tm1-mcp-server@10.5.1在終端機中執行。
{ "mcpServers": { "ffschrattenecker-tm1-mcp-server": { "command": "npx", "args": [ "-y", "@ffschrattenecker/[email protected]" ] } }}claude_desktop_config.json
遠端 URL 還是本機指令? 兩種安裝方式的差異
來自發布者
[!IMPORTANT] This is a personal testing fork, not the project. The author of tm1-mcp-server is flameY3T1. The project to use, cite, link to and report issues against is flameY3T1/tm1-mcp-server, published on npm as .
Personal testing fork of flameY3T1/tm1-mcp-server (npm: tm1-mcp-server). Use the original.
tm1-mcp-serverI (ffschrattenecker) use this fork only to try out changes for myself. It
comes with no support and no stability promise, and it may change, break
or disappear at any time. The scoped package
@ffschrattenecker/tm1-mcp-server exists only so my own machines can
npx it. Its version numbers are counted separately and do not mean it
is newer than the original. Anything worth keeping is offered upstream.
Model Context Protocol (MCP) server for IBM Planning Analytics / TM1. Exposes the full TM1 model lifecycle — metadata, dimensions, cubes, cell I/O, TI processes, chores, security, code-graph analysis — to any MCP-compatible LLM client (Claude Code, Claude Desktop, etc.).
Tested against TM1 11.8 (REST, Basic/CAM auth) and TM1 12.5 / Planning Analytics
Engine (rerooted REST, s2s auth live-validated). Planning Analytics SaaS
(TM1_AUTH_MODE=saas) is supported but was validated upstream only.
Everything below describes this fork's state. For the real project, read the upstream README.
Companion — TM1 IDE for VS Code. Pair this server with the flameY3T1.vscode-tm1-ide extension to browse and inspect the live data, cubes and TI code on the server — an in-place view of whatever MCP just changed.
60 tools across 10 categories — every one listed in docs/TOOLS.md. Past plain CRUD over the REST API:
}StatsByCube runtime evidence),
v12 readiness, orphan dimensions..pro files, or the diff-friendly two-file git layout
({name}.json + {name}.ti).#tm1://process/MyProcess/code attaches a process as context
in chat, plus five starting-point prompts.Token tip: every registered tool's name + input schema costs context on each turn, so exposing all 113 is wasteful when a session needs a slice. Narrow the surface with your client's tool filter;
TM1_MODE=readonlyalready trims it to read-only tools.
To install tm1-mcp-server, follow the
upstream instructions
(npx -y tm1-mcp-server). The steps below install this personal fork. They are
here only for my own setup.
Option A — npm. No clone, no build: npx -y @ffschrattenecker/tm1-mcp-server
always pulls the latest published version. Or install the CLI globally:
npm install -g @ffschrattenecker/tm1-mcp-server
tm1-mcp-server
Option B — clone and build (source / development install):
git clone https://github.com/ffschrattenecker/tm1-mcp-server.git
cd tm1-mcp-server
npm install
npm run build
node dist/index.js
To update: npx picks up new versions on next start (if one is cached,
npm cache clean --force); a global install takes
npm install -g @ffschrattenecker/tm1-mcp-server@latest; a source install takes
git pull && npm install && npm run build. Always restart the MCP client
afterwards — the server process is only spawned at client startup.
Copy .env.example to .env and set the TM1 connection details:
TM1_BASE_URL=https://your-tm1-server:8010
TM1_USER=admin
TM1_PASSWORD=your-password
TM1_SSL_REJECT_UNAUTHORIZED=false
# TM1_PROXY=socks5://proxy:1080 # optional; reach TM1 through a SOCKS5 proxy
TM1_VERSION=11.8
TM1_MODE=readonly # readonly (default) | readwrite
# TM1_ENVIRONMENT=prod # dev | test | prod; prod forces readonly
# TM1_RESPONSE_MODE=structured # legacy (default) | structured
# TM1_MAX_RESPONSE_CHARS=80000 # larger results fail with RESPONSE_TOO_LARGE
# TM1_LOCAL_FILE_ROOT=/srv/tm1-git # optional; confines host-disk file params to this dir
TM1_RESPONSE_MODE=structured drops the text block and ships structuredContent
alone — half the wire bytes, but clients that read content[] (Kiro's IDE among
them) then see nothing. Worth it only for clients that copy the whole result into
the model prompt. .env.example lists every variable with
inline comments; docs/CONFIGURATION.md covers the ones
needing more than a comment line — CAM/LDAP login, the TM1 v12 (Planning
Analytics Engine) connection and its auth modes, host-disk file access.
Safe by default: the server starts in
TM1_MODE=readonly— only read tools are registered, so it cannot mutate or delete anything. SetTM1_MODE=readwriteexplicitly to enable the full lifecycle (cell writes, cube/dimension/process deletion, TI execution), and never point areadwriteserver at production without reviewing the write path first. Label production withTM1_ENVIRONMENT=prod: it then stays readonly even underTM1_MODE=readwrite, unlessTM1_ALLOW_PROD_WRITES=trueis set as well.
Give each connection its own folder with a .env under ~/.tm1/mcp-servers/
(or the folder named by TM1_CONNECTIONS_DIR) and leave TM1_BASE_URL unset:
~/.tm1/mcp-servers/
dev/.env # TM1_BASE_URL=… TM1_MODE=readwrite
prod/.env # TM1_BASE_URL=… (readonly: the default)
One server then serves them all: every tool takes a connection argument and
tm1_list_connections shows what is configured. Each folder sets its own
TM1_MODE — a write against a readonly connection is refused per call — and
no connection logs in before its first use. Folders added or edited later are picked up without
a restart (tm1_list_connections rescans), and the server also starts with none. One entry replaces one MCP server
per TM1 instance, so the client carries one tool list instead of N.
docs/CONFIGURATION.md has the details.
The server holds one TM1 identity per process and every call runs with it. That is the design, not a limitation to work around, and it has consequences worth stating plainly:
TM1_MODE=readwrite with an admin account is a developer configuration,
not a production recommendation. In production, point the server at an account
whose TM1 rights already match what the agent should be able to do — the
server enforces no per-caller authorization of its own.confirm argument repeating the target name
verbatim: tm1_rest_write (every DELETE and cancel action),
tm1_files_write, tm1_delete_elements, tm1_clear_cube,
tm1_execute_process, tm1_execute_chore, tm1_write_cells and
tm1_set_cube_rules; the create-or-update tools ask for it once the target
exists. Anything that can call such a tool can also supply its
confirm value — the guard only stops an auto-approving client firing an
irreversible action without naming the target.If you need multi-user access with per-caller identity, run one process per identity behind your own front end rather than sharing one.
TM1_ALLOW_UNMASKED_SECRETSCredential masking is on by default everywhere it applies. Several tools take a
maskSecrets parameter, and a model can set it to false — an opt-out of a
security control chosen by the thing being guarded against. That request is
ignored unless the operator sets TM1_ALLOW_UNMASKED_SECRETS=true. The default
degrades to "mask anyway" rather than failing the call, so an audit still gets
its report, just redacted.
Copy mcp.json.example to .mcp.json (project-local) or merge it into
~/.claude/settings.json. Do not put TM1_PASSWORD in either file — those
are routinely shared or committed; keep it in a gitignored .env
(where the server looks for one),
or out of files altogether in the OS keychain with
npx -y @ffschrattenecker/tm1-mcp-server secrets migrate <connection>
(details).
For the npx install:
{
"mcpServers": {
"tm1": {
"command": "npx",
"args": ["-y", "@ffschrattenecker/tm1-mcp-server"]
}
}
}
A global install uses "command": "tm1-mcp-server" (no args); a source
build uses "command": "node" with
"args": ["/absolute/path/to/tm1-mcp-server/dist/index.js"], which is what
mcp.json.example shows. Restart Claude Code and the server appears as tm1.
The default transport is stdio, and that is the normal way to run this. For
remote or multi-client setups, TM1_MCP_TRANSPORT=http switches to MCP
Streamable HTTP (stateless, single POST /mcp endpoint). It is single-tenant:
its optional bearer token authenticates the endpoint, not the caller. Setup,
security notes and the autoApprove allowlist:
docs/CONFIGURATION.md.
DataSource.usesUnicode) are dropped when TM1_VERSION says 11.xTM1_INSTANCE + TM1_DATABASE,
live-validated with s2s auth only. tm1_save_data is v11-only. Through
tm1_rest_read/tm1_rest_write: running work is Threads on v11 and Jobs
on v12; v12 serves MessageLogEntries, AuditLogEntries and
TransactionLogEntries empty; v12 dropped demand load, so tm1.Unload has no
successor there;
v12 ships no }Stats* control cubes, so tm1_get_cube_stats and
tm1_audit_feeders in mode runtime/both report the statistics as
unavailable instead of returning metrics (static feeder analysis is unaffected)Names and one-line descriptions: docs/TOOLS.md.
tm1_execute_process and tm1_save_data return an outcome — succeeded,
completed_with_errors, rolled_back or indeterminate — and only succeeded
reports success: true, since TM1 answers HTTP 200 whatever happened.
completed_with_errors means changes were committed: retrying double-posts.
The first connection usually fails on a self-signed dev certificate (set
TM1_SSL_REJECT_UNAUTHORIZED=false — dev only) or a 401 from the wrong
credentials, port or CAM namespace. Those, plus v11/v12 feature errors, slow
transaction-log queries and startup validation errors, are in
docs/CONFIGURATION.md, HTTP-transport
specifics included.
npm run dev # tsx live-reload
npm test # vitest
npm run typecheck # tsc --noEmit
npm run lint:eslint # eslint .
npm run verify # everything CI runs (audit + typecheck + lint gates + tests)
Run npm run verify before opening a PR; it is the gate CI and publish run.
docs/ARCHITECTURE.md covers how to add a tool or service.
autoApproveThis codebase was generated by an AI coding agent (Anthropic's Claude), reviewed and tested but not written by hand. The test suite is also AI-generated and runs against mocks, not a live TM1 server — passing tests do not guarantee correct REST semantics on your Planning Analytics version. Verify behavior against your own instance before relying on it, especially for any write or destructive operation. Provided "as is", best-effort, no support guarantee. See NOTICE.
IBM, Planning Analytics, and TM1 are trademarks of IBM Corp. This project is not affiliated with or endorsed by IBM.
MIT — see LICENSE. The IBM trademarks named above are used solely to describe compatibility with the TM1 REST API.
Personal testing fork of flameY3T1/tm1-mcp-server (npm: tm1-mcp-server). Use the original.
{ "mcpServers": { "ffschrattenecker-tm1-mcp-server": { "command": "npx", "args": [ "-y", "@ffschrattenecker/[email protected]" ] } }}工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
這筆記錄未列出必要的環境變數。
@ffschrattenecker/tm1 透過 stdio 在本機執行。
上次檢查時間:2026-11-11。 Personal testing fork of flameY3T1/tm1-mcp-server (npm: tm1-mcp-server). Use the original.