Fathom MCP 43a42ffb

MCP数据服务

提供方 chand1012

数据服务 stdio · 本地进程 API Key 社区 / 第三方
查看接入方式

Fathom MCP是一个本地优先的会议智能搜索服务器,将Fathom会议转化为可AI搜索的知识库,适用于需要快速、私密、低基础设施的会议智能团队。

一句话选型结论

按资料字段整理

Fathom MCP 43a42ffb 属于数据服务类 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。

分类数据服务 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 5 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

数据服务
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5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

会议搜索Python本地部署本地AI知识库SQLite向量搜索

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
uv run main.py

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 Python,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 5 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
数据库查询

用自然语言查库、取数,不用手写 SQL。

数据同步与清洗

把多来源数据拉到一处做整理。

报表取数

按口径快速拉出统计结果。

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 5 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

  • 需要在数据服务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
Fathom MCP 43a42ffb当前 5 stdio · 本地进程 本地部署 有 2026-05-17
Memory 9d25f64e 9 未标注 未标注 无 2026-05-17
Getzep MCP 9 stdio · 本地进程 本地部署 有 2026-05-17
anything-llm 未标注 stdio · 本地进程 本地部署 有 2026-04-11
Prisma MCP Server 未标注 stdio · 本地进程 本地部署 有 2026-05-17
Astr Bot 未标注 stdio · 本地进程 本地部署 有 2026-05-17

服务器档案

数据服务
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chand1012上游标注的维护方 chand1012
分类数据服务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 · Python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

9.6 KB
来自项目仓库与上游目录的原文,未做删改

Fathom has an official MCP now, you should juse use that one.

Fathom MCP

A local-first MCP server that turns Fathom meetings into an AI-searchable knowledge base.

Fathom MCP ingests meetings and transcripts from Fathom, generates embeddings locally with llama.cpp, stores everything in SQLite + sqlite-vec, and exposes search/retrieval tools through FastMCP.

It is built for teams that want fast, private-by-default, low-infrastructure meeting intelligence that stays on their machine and works with MCP-native clients.


Why This Project

Meeting recordings are high-value, but hard to query quickly.

This server gives you:

  • Webhook and sync ingestion from Fathom.
  • Semantic transcript search across all meetings.
  • In-meeting transcript drilling for focused QA.
  • A local vector stack (SQLite + GGUF embeddings) instead of external vector infrastructure.
  • MCP tools that can be consumed by AI assistants and agent workflows.

Local-First, By Design

Fathom MCP is intentionally designed so your searchable meeting knowledge lives locally:

  • Transcript chunks and metadata are stored in local SQLite.
  • Embeddings are generated in-process with a local GGUF model via llama.cpp.
  • Similarity search runs with sqlite-vec in the same local database.
  • No external vector DB, no hosted retrieval service, no extra data pipeline.

The only network dependency is Fathom ingestion (API/webhooks) and the one-time model download if the GGUF file is not already present.


Core Capabilities

  • Hybrid server: FastAPI API endpoints + FastMCP endpoint in one ASGI app.
  • Authenticated MCP: Bearer token auth backed by constant-time comparison.
  • Webhook validation: Svix-style signature verification with replay protection.
  • Background processing: Webhook and sync jobs return quickly and process async.
  • Vector search: sqlite-vec cosine distance search over locally stored transcript embeddings.
  • Local embedding model: Automatic GGUF download once, then local reuse at startup.
  • Minimal moving parts: No external cache, broker, or vector service required.

Architecture```mermaid

flowchart LR
A[Fathom API + Webhooks] --> B[FastAPI Router]
B --> C[Meeting Service]
C --> D[(SQLite meetings + transcripts)]
C --> E[llama.cpp Embedder]
E --> D
D --> F[sqlite-vec similarity search]
F --> G[FastMCP Tools /mcp/]
D --> H[HTTP Tool Endpoints /tools/*]

subgraph L[Local-First Runtime]
  B
  C
  D
  E
  F
  G
  H
end

---

## Quick Start

### 1) Prerequisites

- Python 3.14+
- `uv`
- A Fathom API key and webhook secret

Install `uv` if needed:```bash
curl -LsSf https://astral.sh/uv/install.sh | sh

2) Install dependencies```bash

uv sync


### 3) Configure environment

Create a `.env` file in the project root:```env
# Required
FATHOM_API_URL=https://api.fathom.ai/external/v1
FATHOM_API_KEY=your_fathom_api_key
FATHOM_WEBHOOK_SECRET=whsec_base64_encoded_secret
SERVICE_API_KEY=generate_a_unique_random_secret_for_this_server

# Optional (defaults shown)
HOST=0.0.0.0
PORT=8000
VECTOR_DB_PATH=./data/vectors.db
EMBEDDING_DIMENSION=384
EMBEDDING_MODEL_PATH=./data/bge-small-en-v1.5-q8_0.gguf
EMBEDDING_MODEL_URL=https://huggingface.co/ggml-org/bge-small-en-v1.5-Q8_0-GGUF/resolve/main/bge-small-en-v1.5-q8_0.gguf

Notes:

  • Environment names are case-insensitive in this project configuration.
  • SERVICE_API_KEY must be a real secret, not a placeholder such as your_service_api_key.
  • On first startup, the embedding model is downloaded automatically if missing.
  • If EMBEDDING_MODEL_PATH already exists, startup stays fully local for embedding operations.

4) Run the server```bash

uv run main.py


For development reload:```bash
uv run --reload main.py

Server defaults to http://localhost:8000.

5) Configure the Fathom webhook (recommended)

To receive near real-time meeting updates, configure Fathom to send newMeeting
events to this server.

Webhook endpoint:```text
POST /webhook


Requirements:

- Your endpoint must be publicly accessible from the internet.
- Your endpoint must be served over HTTPS (valid TLS certificate).
- The webhook secret configured in Fathom must match `FATHOM_WEBHOOK_SECRET`.

Typical production URL example:```text
https://your-domain.com/webhook

If running locally, expose your service through an HTTPS tunnel/reverse proxy
(for example: Cloudflare Tunnel, ngrok, or a public ingress) and use that
public HTTPS URL in Fathom.


API and MCP Endpoints

Public routes

  • POST /webhook - receive Fathom newMeeting webhook payloads.
  • POST /sync - trigger background pull of recent meetings.
  • GET /tools/* - HTTP mirror of MCP tools for quick testing.
  • POST|GET /mcp/ - FastMCP streamable HTTP endpoint.

Auth model

  • /sync, /tools/*, and /mcp/ require the shared service API key middleware:
  • Authorization: Bearer
  • X-API-Key:
  • /webhook requires valid Svix-style signature headers:
  • webhook-id
  • webhook-timestamp
  • webhook-signature
  • /docs, /redoc, and /openapi.json are public (unauthenticated)

Example calls

Trigger sync:```bash
curl -X POST http://localhost:8000/sync \
-H "Authorization: Bearer ${SERVICE_API_KEY}"

Equivalent FastAPI auth header:

curl -X POST http://localhost:8000/sync \
-H "X-API-Key: ${SERVICE_API_KEY}"


Search meetings:```bash
curl "http://localhost:8000/tools/search_meetings?title=roadmap&limit=5" \
  -H "Authorization: Bearer ${SERVICE_API_KEY}"

Search transcripts semantically:```bash
curl "http://localhost:8000/tools/search_transcripts?query=What%20did%20we%20decide%20about%20pricing%3F&limit=5" \
-H "Authorization: Bearer ${SERVICE_API_KEY}"


Search within one meeting:```bash
curl "http://localhost:8000/tools/search_meeting_transcripts?meeting_id=123&query=action%20items&limit=5" \
  -H "Authorization: Bearer ${SERVICE_API_KEY}"

Get meeting by ID:```bash
curl "http://localhost:8000/tools/get_meeting?meeting_id=123" \
-H "Authorization: Bearer ${SERVICE_API_KEY}"


Get full transcript chunks for a meeting:```bash
curl "http://localhost:8000/tools/get_meeting_transcript?meeting_id=123" \
  -H "Authorization: Bearer ${SERVICE_API_KEY}"

MCP Tools Exposed

The FastMCP server exposes these tool functions at /mcp for AI assistants and agent workflows:

  • search_meetings(title, start_date, end_date, limit)
  • search_transcripts(query, limit)
  • search_meeting_transcripts(meeting_id, query, limit)
  • get_meeting(meeting_id)
  • get_meeting_transcript(meeting_id)

These map to the same data access layer used by the HTTP /tools/* routes.


Local Data Model

SQLite database tables:

  • meetings: canonical meeting metadata and summary/action-items payloads.
  • transcripts: transcript chunks with associated embedding blob.
  • vec_transcripts (virtual): vector index for KNN/cosine search via sqlite-vec.

Default DB path: ./data/vectors.db.

This file-centric layout makes backup, migration, and inspection straightforward.


Docker

Published image:```bash
ghcr.io/chand1012/fathom-mcp:latest


Pull the latest image from GHCR:```bash
docker pull ghcr.io/chand1012/fathom-mcp:latest

Run container:```bash
mkdir -p ./data

docker run --rm -p 8000:8000 \
-v "$(pwd)/data:/data" \
-e FATHOM_API_URL="https://api.fathom.video" \
-e FATHOM_API_KEY="your_fathom_api_key" \
-e FATHOM_WEBHOOK_SECRET="whsec_base64_encoded_secret" \
-e SERVICE_API_KEY="generate_a_unique_random_secret_for_this_server" \
ghcr.io/chand1012/fathom-mcp:latest


If you prefer files on the host to be owned by your current user, add:```bash
-u "$(id -u):$(id -g)"

The ./data directory on your host is mounted to /data in the container,
so both /data/bge-small-en-v1.5-q8_0.gguf and /data/vectors.db persist
between container restarts.

Optional: if you want to build locally instead of using GHCR:```bash
docker build -t fathom-mcp .


---

## Development Workflow

Run tests:```bash
uv run pytest

Run specific test:```bash
uv run pytest test_full_workflow.py::test_full_workflow


Lint and format:```bash
uv run ruff check .
uv run ruff format .

Type checking:```bash
uv run mypy .


---

## Project Layout```text
src/fathom_mcp/
  api/
    client.py      # Fathom API client
    router.py      # Webhook/sync/tool HTTP endpoints
    service.py     # Fetch + store + transcript vectorization workflow
  core/
    config.py      # Pydantic settings and environment configuration
  mcp/
    server.py      # FastMCP server and tool registrations
  vector/
    database.py    # SQLite + sqlite-vec storage/search layer
    embedder.py    # Local llama.cpp embedding model integration
  webhooks/
    handler.py     # Svix signature verification + webhook ingestion
main.py            # Application assembly and startup lifecycle

Security Notes

  • Do not commit .env or secrets.
  • Rotate SERVICE_API_KEY and webhook secret periodically.
  • The middleware protects the API and mounted MCP app with the same service key.
  • Validate webhook signatures before processing payloads (already enforced).
  • Local-first reduces external data exposure by default because storage and vector search are on-box.

Roadmap Ideas

  • Incremental sync scheduling.
  • Multi-tenant key management.
  • Optional cloud embedding providers.
  • Richer MCP tool metadata and pagination support.
  • Observability dashboards for ingest/search latency.

Contributing

PRs are welcome. Keep changes focused, include tests for behavior changes, and run lint/type checks before opening a PR.


License

No license file is currently included in this repository.

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
Fathom MCP 43a42ffb 怎么安装?

在客户端按下面的安装命令接入:uv run main.py。装完在客户端配置里引用同一个包或端点。

Fathom MCP 43a42ffb 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。

Fathom MCP 43a42ffb 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。

Fathom MCP 43a42ffb 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。

Fathom MCP 43a42ffb 能提供哪些能力?

资料显示约 5 个工具;能力标签包括 本地AI、Python、向量搜索、知识管理。具体工具清单以官方文档为准。

Fathom MCP 43a42ffb 需要什么运行环境?

运行时为 Python;属于本地进程,需要在本机装好运行环境。

Fathom MCP 43a42ffb 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。