Knowledge MCP B171fc50

MCP搜索检索

提供方 srclos-net

搜索检索 stdio · 本地进程 无需鉴权 社区 / 第三方
查看接入方式

一个自托管的MCP服务器,为AI助手提供跨对话的持久记忆,通过存储和检索学习内容来优化每次对话。

一句话选型结论

按资料字段整理

Knowledge MCP B171fc50 属于搜索检索类 MCP 服务器,stdio · 本地进程,无需鉴权,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。

分类搜索检索 传输方式stdio · 本地进程 鉴权无需鉴权 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 6 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

搜索检索
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6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

AI记忆Go本地部署上下文管理自托管语义搜索SQLiteChromaDB

安装与配置

stdio · 本地进程 无需鉴权
客户端在本机拉起子进程,数据不出本机
docker run -d \

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 Docker,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 6 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
联网搜索取证

让模型查最新资料并给出出处。

资料聚合

把一个主题的网页、文档集中抓回来整理。

情报与竞品检索

按关键词持续检索行业动态和竞品信息。

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 6 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。
  • 资料标注无需鉴权,装好就能连。

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。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在搜索检索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。
  • 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
Knowledge MCP B171fc50当前 6 stdio · 本地进程 本地部署 有 2026-05-17
Brave Search 433812ff 2 未标注 未标注 无 2026-05-17
Sequentialthinking 1 SSE · 服务端推送 远程部署 有 2026-05-18
Baidu Netdisk 未标注 未标注 未标注 无 2026-05-17
ragflow 未标注 stdio · 本地进程 本地部署 有 2026-04-11
mindsdb (430e18) 未标注 未标注 未标注 有 2026-05-17

服务器档案

搜索检索
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srclos-net上游标注的维护方 srclos-net
分类搜索检索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
鉴权方式无需鉴权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 · Docker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

9.0 KB
来自项目仓库与上游目录的原文,未做删改

knowledge-mcp

A self-hosted MCP (Model Context Protocol) server that gives AI assistants persistent memory across conversations. The AI looks up relevant context at the start of each session and writes back learnings it discovers — preferences, mistakes to avoid, personal context, communication patterns — so every future conversation is better informed than the last.

Built in Go. Stores data in either SQLite (simple, self-contained) or ChromaDB (semantic search, existing infrastructure). Speaks the MCP streamable HTTP transport, compatible with open-webui and other MCP clients.


How it works

At the start of a conversation the AI calls lookup_context with keywords describing what you're working on. The server searches stored learnings and returns relevant ones — things like your communication preferences, past mistakes the AI made, relevant personal context, or technical details about your stack. The AI uses this to calibrate its response before saying a word.

At the end of a session (or whenever something useful is discovered), the AI calls store_learning to persist it. Over time the store builds up a detailed, searchable picture of how to work with you effectively.

The AI writes directly — no human approval step. You can review, edit, or prune entries at any time using list_learnings, update_learning, and delete_learning.


Tools exposed

Tool Description
lookup_context Call this first. Searches stored learnings by keyword and returns relevant ones. Increments use count on returned results.
store_learning Stores a new learning with category, content, tags, and confidence score.
list_learnings Lists all stored learnings, optionally filtered by category.
update_learning Updates an existing learning by ID.
delete_learning Deletes a learning by ID.
get_stats Returns a count of learnings per category.

Categories

Category Purpose
preferences Communication style, formatting preferences, things to always/never do
personal_context Relevant personal facts that inform better responses
technical Stack details, tools in use, technical preferences
personal_growth Ongoing work, patterns, things the person is working through
mistakes Things that went wrong and how to avoid repeating them
general Catch-all for anything that doesn't fit above

Backends

SQLite (default)

Simple, self-contained. A single file on disk. Uses FTS5 full-text search when available (Alpine Linux includes it), falls back to per-word LIKE queries otherwise. Good for single-instance deployments.

The database is created automatically on first run — no setup needed beyond ensuring the directory exists.

ChromaDB

Uses your existing ChromaDB HTTP API (v2). Stores documents with metadata. If you configure an Ollama embedding model, embeddings are generated via Ollama and passed to Chroma, giving you real semantic search rather than keyword matching. Without an embedding model configured, Chroma uses its own default embedder.

Requires ChromaDB ≥ 0.6 (API v2).


Configuration

Configuration is a TOML file. Print an example with:```bash
./self-improvement-mcp --print-config


### Full reference```toml
[server]
addr = ":8080"          # Listen address

[backend]
type = "sqlite"         # "sqlite" or "chroma"

[sqlite]
path = "/data/learnings.db"   # Path to SQLite file; created automatically

[chroma]
url        = "http://chroma:8000"
tenant     = "default_tenant"    # Chroma v2 tenant (default: "default_tenant")
database   = "default_database"  # Chroma v2 database (default: "default_database")
collection = "self_improvement"  # Collection name; created automatically if missing

# Optional: use Ollama for semantic embeddings
# embedding_model = "nomic-embed-text"
# ollama_url      = "http://ollama:11434"

Config file resolution order

The server looks for a config file in this order, stopping at the first one found:

  1. --config flag
  2. CONFIG_FILE environment variable
  3. ./config.toml (current directory)
  4. /config/config.toml
  5. /etc/self-improvement-mcp/config.toml

If no config file is found, all defaults apply (SQLite backend, port 8080, /data/learnings.db).


Running locally

Build from source

Requires Go 1.22+ and a C compiler (for SQLite CGO bindings).```bash
git clone <repo>
cd self-improvement-mcp
go build -o self-improvement-mcp .
./self-improvement-mcp --print-config > config.toml

edit config.toml as needed

./self-improvement-mcp --config config.toml


### Run with SQLite```bash
mkdir -p data
./self-improvement-mcp --config config.toml

The SQLite database is created automatically. No touch or pre-initialization needed.

Run with ChromaDB

Update config.toml:```toml
[backend]
type = "chroma"

[chroma]
url = "http://192.168.1.20:8001"
collection = "self_improvement"


The collection is created automatically on first run if it doesn&#x27;t exist.

---

## Running with Docker

### Build the image```bash
docker build -t self-improvement-mcp:latest .

The Dockerfile uses a multi-stage Alpine build. The final image is ~15MB.

Run with SQLite```bash

docker run -d \
--name self-improvement-mcp \
-p 8080:8080 \
-v $(pwd)/config.toml:/config/config.toml:ro \
-v self-improvement-data:/data \
self-improvement-mcp:latest


### Run with ChromaDB```bash
docker run -d \
  --name self-improvement-mcp \
  -p 8080:8080 \
  -v $(pwd)/config.toml:/config/config.toml:ro \
  self-improvement-mcp:latest

Connecting from another container

If your MCP client (e.g. open-webui) runs in a container and the server runs on the Docker host, use:```
http://host.docker.internal:8080/mcp


On Linux, `host.docker.internal` requires explicit configuration. Add to your client container&#x27;s compose service:```yaml
extra_hosts:
  - &quot;host.docker.internal:host-gateway&quot;

Kubernetes deployment

A complete manifest is provided in k8s.yaml. It includes a ConfigMap, PVC (for SQLite), Deployment, and Service, all in the ai namespace.

Deploy```bash

Update the image reference in k8s.yaml first

kubectl apply -f k8s.yaml


### Switch backends via ConfigMap

To switch from SQLite to ChromaDB, edit the ConfigMap and update `type = &quot;chroma&quot;` — no rebuild or redeploy needed, just a rolling restart:```bash
kubectl edit configmap self-improvement-mcp-config -n ai
kubectl rollout restart deployment/self-improvement-mcp -n ai

Service URL (from within the cluster)```

http://self-improvement-mcp.ai.svc.cluster.local:8080/mcp


---

## Connecting to open-webui

In open-webui, add a new MCP tool server:

- **URL:** `http://host.docker.internal:8080/mcp` (if running on Docker host)
- **Transport:** Streamable HTTP

The server uses the MCP `instructions` field in the `initialize` response to tell the AI to call `lookup_context` at the start of every conversation. How strictly this is followed depends on the model and open-webui&#x27;s system prompt configuration.

---

## Endpoints

| Endpoint | Method | Description |
|----------|--------|-------------|
| `/mcp` | `POST` | MCP JSON-RPC endpoint (streamable HTTP) |
| `/mcp` | `GET` | SSE stream for server-initiated messages |
| `/health` | `GET` | Health check — returns `{&quot;status&quot;:&quot;ok&quot;,&quot;version&quot;:&quot;1.0.0&quot;}` |

---

## MCP protocol

The server implements MCP 2024-11-05 over streamable HTTP. Supported methods:

- `initialize`
- `notifications/initialized`
- `ping`
- `tools/list`
- `tools/call`
- Batch requests (JSON array)

---

## Project structure```
self-improvement-mcp/
├── main.go              # Entry point, config loading, server startup
├── config.go            # TOML config types and defaults
├── backend.go           # Backend interface + factory
├── backend_sqlite.go    # SQLite implementation (FTS5 + LIKE fallback)
├── backend_chroma.go    # ChromaDB v2 HTTP API implementation
├── server.go            # Streamable HTTP MCP server
├── tools.go             # Tool definitions and handlers
├── Dockerfile           # Multi-stage Alpine build
└── k8s.yaml             # Kubernetes manifests (ConfigMap, PVC, Deployment, Service)

Adding a new backend

Implement the Backend interface in backend.go:go type Backend interface { Add(category, content, tags string, confidence float64) (*Learning, error) Search(query, category string, limit int) ([]*Learning, error) List(category string, limit int) ([]*Learning, error) Update(id, content, tags string, confidence float64) error Delete(id string) error IncrementUseCount(id string) Stats() (map[string]int, error) Close() error }

Then add a case to the NewBackend factory in backend.go and a new config section in config.go.


Dependencies

Dependency Purpose
github.com/mattn/go-sqlite3 SQLite driver (CGO)
github.com/BurntSushi/toml TOML config parsing

No other runtime dependencies. ChromaDB and Ollama are external services, not Go dependencies.


License

MIT

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
Knowledge MCP B171fc50 怎么安装?

在客户端按下面的安装命令接入:docker run -d \。装完在客户端配置里引用同一个包或端点。

Knowledge MCP B171fc50 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。

Knowledge MCP B171fc50 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。

Knowledge MCP B171fc50 需要 API Key 吗?

资料标注为无需认证,装好就能连;如果实际调用时报鉴权错误,按官方文档补上凭据。

Knowledge MCP B171fc50 能提供哪些能力?

资料显示约 6 个工具;能力标签包括 AI记忆、上下文管理、Go、搜索、自托管。具体工具清单以官方文档为准。

Knowledge MCP B171fc50 需要什么运行环境?

运行时为 Docker;属于本地进程,需要在本机装好运行环境。

Knowledge MCP B171fc50 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。