MCP Gateway 789a2f02

MCPAI代理

提供方 GaussGeorge

AI代理 stdio · 本地进程 API Key 社区 / 第三方
查看接入方式

PlanGate是一个MCP兼容的网关,通过多步会话语义对齐准入决策,减少多步LLM代理工作负载中的治理引起的级联计算浪费。

一句话选型结论

按资料字段整理

MCP Gateway 789a2f02 属于AI代理类 MCP 服务器,stdio · 本地进程,API Key,有可直接执行的安装命令。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类AI代理 传输方式stdio · 本地进程 鉴权API Key 部署方式本地部署
GitHub 星标 0 仓库热度,无公开仓库时显示未标注
工具数 7 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

AI代理
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量7暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

适用场景

LLM代理Python本地部署会话治理计算优化多步工作流准入控制

安装与配置

stdio · 本地进程 API Key
客户端在本机拉起子进程,数据不出本机
pip install -r mcp_server/requirements.txt

这条命令只是入口,具体启动参数请按官方文档补齐,不要直接照搬到别的服务器。

  1. 1
    准备运行环境

    stdio 服务在本地起进程,需要先装好 Node.js 等运行时,再按安装命令拉取依赖。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后客户端会列出约 7 个工具,随便调一个只读工具确认能返回结果。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
给代理接外部能力

让代理在任务中途调用搜索、数据库等真实工具。

多步任务编排

把一串操作拆成步骤,由代理按顺序执行。

自动化重复流程

把固定流程交给代理跑,减少手工操作。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 资料标注 7 个工具,职责相对集中,接入后不容易误操作。
  • 有现成的安装命令,不用自己找入口,照着跑就能起服务。

先想清楚的地方

  • stdio 方式要求本地装好运行时(如 Node.js),换机器要重新装一遍。
  • 使用前要准备好API Key对应的凭据,并按最小权限配置。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在AI代理场景里让模型直接调用外部能力的人。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 完全不想在本地装环境的人。
  • 不方便管理密钥或第三方授权的人。
  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
MCP Gateway 789a2f02当前 7 stdio · 本地进程 本地部署 有 2026-05-17
Filesystem 560c4363 13 未标注 本地部署 无 2026-05-17
Server Everything (everything) 11 stdio · 本地进程 本地部署 有 2026-05-17
Time 7365cb6d 2 SSE · 服务端推送 远程部署 有 2026-05-18
Everything Claude Code 未标注 stdio · 本地进程 本地部署 有 2026-05-17
Openmemory 557e1585 未标注 stdio · 本地进程 本地部署 有 2026-05-17

服务器档案

AI代理
字段来自项目仓库与公开索引,中文说明由本站标注
提供方GaussGeorge上游标注的维护方 GaussGeorge
分类AI代理本站按上游分类展示
连接方式stdio · 本地进程客户端在本机拉起子进程,数据不出本机
鉴权方式API Key远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时未标注 · 未标注本地安装时用到的包标识与运行时
部署类型本地部署上游给出的部署形态
最近核验2026-05-17上游最后核验时间

详细介绍

9.7 KB
来自项目仓库与上游目录的原文,未做删改

# PlanGate: Session Commitment for Multi-Step LLM Agent Tool Governance

PlanGate is an MCP-compatible gateway that reduces governance-induced cascading compute waste
in multi-step LLM agent workloads by aligning admission decisions with multi-step session
semantics — enforcing session commitment: atomic admission, temporal isolation, and
continuation value.

For plan-accurate P&S sessions, PlanGate prevents price-induced mid-session rejection
within the declared plan scope. For ReAct sessions, it reduces mid-execution abandonment
through sunk-cost-aware continuation pricing.

Agent type Commitment Mechanism
Plan-and-Solve (P&S) Budget-reserved Pre-flight DAG admission + locked price
ReAct Soft continuation incentive K^2 sunk-cost discount pricing

Additionally, PlanGate-R is a checkpoint-aware recovery extension evaluated
in a controlled mock runtime (P&S only; no real LLM; no ReAct semantic
recovery). It is not a primary contribution.


Repository Structure```

cmd/gateway/ Go gateway entry point (main.go)
plangate/ Core Go package: admission, pricing, DAG validation,
checkpoint/recovery, reputation
baseline/ Baseline gateway implementations (NG, SRL, SBAC,
DAGOR, Rajomon, ProgressPriority)
mcp_server/ Python MCP backend + tool implementations
server.py ThreadPoolExecutor MCP server
tools/ calculator, weather, web_fetch, mock_heavy,
llm_reasoner, deepseek_llm, ...
scripts/ Experiment runner, analysis, and plotting scripts
run_all_experiments.py Main mock experiment driver
run_exp_real3_all.sh Real-LLM (GLM/DeepSeek) experiment driver
reproduce_mock_core.sh Reproducibility: mock core experiments
reproduce_main_paper_from_cache.sh Re-generate figures from CSV cache
gen_paper_figures.py Paper figure generator
docs/ Artifact documentation
REPRODUCIBILITY.md Full reproduction guide with expected runtimes
RESULT_MAPPING.md Maps paper items to data files
experiment_code_mapping.md


---

## Requirements

| Requirement | Version | Notes |
|---|---|---|
| Go | >= 1.23 | `go version` |
| Python | >= 3.10 | `python --version` |

Install Python dependencies:```bash
pip install -r mcp_server/requirements.txt

Build the gateway binary:```bash
go build -o gateway ./cmd/gateway

Cross-compile for Linux:

GOOS=linux GOARCH=amd64 CGO_ENABLED=0 go build -o gateway_linux ./cmd/gateway


---

## Quick Start: Smoke Test (no API keys required)

All unit tests and mock experiments run without any real LLM provider.```bash
# Run full Go test suite
go test ./... -timeout 120s

# Run plangate package tests only
go test ./plangate/... -timeout 120s -v

Expected output: ok mcp-governance/plangate, ok mcp-governance/baseline,
ok mcp-governance.

Minimal end-to-end mock smoke test:```bash

1. Start the Python mock backend

python mcp_server/server.py --host 127.0.0.1 --port 8080 --max-workers 10 &

2. Start the PlanGate gateway in mock mode

./gateway --mode mcpdp --backend http://127.0.0.1:8080 --port 9200 \
--max-sessions 200 --base-price 10 --alpha 0.5 &

3. Smoke test (50 sessions, no API key needed)

python scripts/smoke_test_multitool.py \
--gateway http://127.0.0.1:9200 --sessions 50


---

## Running Controlled Mock Experiments (Tier 1)```bash
# All 12 mock experiments, 5 repeats each
python scripts/run_all_experiments.py --exp all --repeats 5

# Single experiment
python scripts/run_all_experiments.py --exp Exp1_Core --repeats 5

# With CPU isolation (Linux/WSL2)
python scripts/run_all_experiments.py --exp all --repeats 5 \
    --gateway-binary ./gateway_linux \
    --cpu-backend 8-15 --cpu-gateway 4-7 --cpu-loadgen 0-3

Results are written to results/exp1_core/, results/exp2_heavyratio/, etc.
The results/ directory is excluded from the public repo (see .gitignore).


Reproducing Key Results

The public repository contains source code and no-key minimal reproduction commands.
Full cached traces for real-LLM experiments are not tracked in this public branch;
they are distributed separately through the conference artifact submission mechanism,
not as a public GitHub release.

Mock core experiments can be re-run from scratch:```bash

Re-run mock core experiments from scratch (no API key, ~30–45 min)

bash scripts/reproduce_mock_core.sh

Re-generate paper figures from local CSV results (after re-running)

python scripts/gen_paper_figures.py


See docs/REPRODUCIBILITY.md for full instructions,
expected runtimes, and troubleshooting.
See docs/RESULT_MAPPING.md for a mapping of paper
items to specific data files.

---

## Running Real-LLM Experiments (Tier 2 / 3)

Real-LLM experiments require a provider API key. Copy `.env.example` to `.env`
and fill in `LLM_API_KEY`:```bash
cp .env.example .env
# Edit .env and set LLM_API_KEY, LLM_API_BASE, LLM_MODEL
``````bash
# Tier 2: Steady real-LLM (GLM-4-Flash or DeepSeek-V3)
bash scripts/run_exp_real3_all.sh                    # default: GLM-4-Flash
bash scripts/run_exp_real3_all.sh --deepseek         # DeepSeek-V3

# Tier 3: Bursty real-LLM
python scripts/run_real_llm_bursty.py --repeats 3 --burst-size 30

Gateway Strategies

Strategy Mode flag Description Source
No-Gov (NG) ng Pass-through, no governance baseline/ng_gateway.go
SRL srl Static rate limiter (token bucket) baseline/srl_gateway.go
SBAC sbac Session-count-based admission baseline/sbac_gateway.go
DAGOR dagor RTT-threshold overload detection baseline/dagor_gateway.go
Rajomon rajomon Queuing-delay dynamic pricing baseline/rajomon_gateway.go
ProgressPriority pp Step-progress priority baseline/progress_priority_gateway.go
PlanGate mcpdp Session commitment (this work) plangate/, cmd/gateway/

PlanGate-R: Checkpoint-Aware Recovery Extension

PlanGate-R handles recoverable transient backend failures within admitted P&S
sessions by checkpointing completed steps and resuming from the last verified
step, rather than restarting from scratch.

Scope: Evaluated in a controlled mock runtime only (P&S sessions, mock
handlers, injected recoverable failures, no real LLM). Not a ReAct semantic
recovery experiment. Results do not transfer to real-LLM or production
workloads without further evaluation.

Key result at failure_rate=0.3: PlanGate-R matches naive retry's eventual
success rate (1.000) while reducing waste steps by 66.7% and total tool calls
by 10.1% per run.


Minimal Reproduction

> Mock reproduction does not require API keys.
> PlanGate-R reproduction is P&S controlled mock only.
> Exact paper tables require full runs; minimal smoke validates qualitative trends.
> See docs/minimal_reproduction.md and
> docs/paper_mapping.md for full details.

1. Unit tests (no server, no API key, < 1 min):```bash
go test ./... -timeout 120s


**2. PlanGate core mock smoke — validates PlanGate reduces cascade vs NG/SBAC (no API key, ~1–5 min, verified 1.2 min on Windows):**```bash
# Linux / macOS / WSL2
go build -o gateway ./cmd/gateway
python scripts/run_all_experiments.py --exp Exp1_Core --repeats 1 --gateway-binary ./gateway

# Windows PowerShell
go build -o gateway.exe ./cmd/gateway
python scripts/run_all_experiments.py --exp Exp1_Core --repeats 1 --gateway-binary gateway.exe

> The gateway binary is built locally and excluded from git. Omit --gateway-binary to let the script auto-build.

Sanity check: plangate_full.cascade_failed == 0, plangate_full.effective_goodput is highest.

3. PlanGate mechanism ablation smoke — validates budget-lock matters (no API key, ~1–3 min, verified 0.8 min on Windows):```bash

Linux / macOS / WSL2

python scripts/run_all_experiments.py --exp Exp4_Ablation --repeats 1 --gateway-binary ./gateway

Windows PowerShell

python scripts/run_all_experiments.py --exp Exp4_Ablation --repeats 1 --gateway-binary gateway.exe

Sanity check: `wo_budgetlock.effective_goodput` ~83% lower than `plangate_full`.

**4. PlanGate-R recovery smoke — validates checkpoint resume / no replay (no API key, &lt; 2 min, Go tests only):**```bash
go test ./plangate/... -run &quot;TestRuntime&quot; -v -timeout 120s

What is NOT included:
- paper/ — unpublished draft, excluded entirely.
- results/ — large experiment CSVs, excluded (see .gitignore).
- Real LLM credentials — not needed for mock or unit tests.
- Large tokenizer asset (scripts/deepseek_v3_tokenizer/tokenizer.json) — see
scripts/deepseek_v3_tokenizer/README.md.
- Real LLM / vLLM experiments (Tables 6–8) are optional and require external setup.


Artifact Notes

  • paper/ is excluded from this public repository.
  • results/ (large experiment CSVs) is excluded from this public repository.
  • API keys: copy .env.example to .env and fill in values.
    Mock experiments and all unit tests do not require any API key.
  • Large external tokenizer assets are not tracked; see
    scripts/deepseek_v3_tokenizer/README.md
    for how to obtain tokenizer.json if needed for token-accounting scripts.
  • Pre-computed cached CSV traces for real-LLM experiments are not included in
    this public repository. For double-blind submission, they are provided through
    the conference supplementary artifact mechanism.

License

This repository is released under the Apache License 2.0. See LICENSE for details.


Citation```bibtex

@inproceedings{plangate2026,
title = {PlanGate: Session Commitment for Multi-Step LLM Agent Tool Governance},
author = {Anonymous},
booktitle = {Anonymous Submission},
year = {2026}
}
```

常见问题

7 问
答案由本站依据条目字段整理,不引用上游原文
MCP Gateway 789a2f02 怎么安装?

在客户端按下面的安装命令接入:pip install -r mcp_server/requirements.txt。装完在客户端配置里引用同一个包或端点。

MCP Gateway 789a2f02 怎么在客户端里配置?

它的传输方式未标注,在客户端 MCP 配置里按官方文档填写启动命令和参数,保存后重启客户端生效。

MCP Gateway 789a2f02 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

MCP Gateway 789a2f02 需要 API Key 吗?

资料标注需要API Key认证,使用前准备好对应凭据,并按最小权限配置。

MCP Gateway 789a2f02 能提供哪些能力?

资料显示约 7 个工具;能力标签包括 LLM代理、Python、AI代理。具体工具清单以官方文档为准。

MCP Gateway 789a2f02 需要什么运行环境?

属于本地进程,需要在本机装好运行环境。

MCP Gateway 789a2f02 是本地还是远程?

它是本地进程(stdio),由客户端在本机拉起。传输方式为stdio · 本地进程。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 2026-05-17。