Overview

ChatSum is an MCP (Model Context Protocol) server that produces concise, LLM-powered summaries of chat histories. It implements MCP-compatible endpoints so other tools and agents can push conversation context and receive condensed insights, highlights, or queryable summaries. The server is intended to make large, multi-turn chat logs easier to browse, search, and export for downstream workflows.

By centralizing summarization and export logic, ChatSum helps teams extract the important points from long conversations — meeting notes, support chats, design discussions, or code review threads — without manually sifting through the full transcript. It supports configurable summary formats, export targets, and can be plugged into an existing MCP network or used as a standalone service.

Repository: https://github.com/mcpso/mcp-server-chatsum

Features

Summarize chat transcripts into short, medium, or long summaries
Queryable summaries: ask questions about the conversation and receive context-aware answers
Multiple export formats: JSON, Markdown, and plain text
Configurable prompt templates and summary schemas
Streaming support for progressive result delivery (where supported by the LLM backend)
Cache and persistence options for reuse and deduplication
Authentication and basic access controls to protect conversation data
MCP-compatible API so it integrates with model orchestrators and clients

Installation / Configuration

Prerequisites:

Docker (recommended) or Python 3.10+
API keys for one or more LLM providers (e.g., OpenAI, Anthropic) if using external models

Quickstart with Docker Compose:

# docker-compose.yml
version: "3.8"
services:
  chatsum:
    image: mcpso/mcp-server-chatsum:latest
    ports:
      - "8080:8080"
    environment:
      - MCP_PORT=8080
      - LLM_PROVIDER=openai
      - OPENAI_API_KEY=${OPENAI_API_KEY}
    volumes:
      - ./data:/app/data

Run:

OPENAI_API_KEY="sk-..." docker compose up -d

Local Python install (development):

git clone https://github.com/mcpso/mcp-server-chatsum.git
cd mcp-server-chatsum
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
export OPENAI_API_KEY="sk-..."
python ./server.py --port 8080

Example configuration file (config.yml):

server:
  host: 0.0.0.0
  port: 8080

llm:
  provider: openai
  model: gpt-4o
  max_tokens: 800

summaries:
  default_length: short
  include_questions: true
  export_formats: ["json", "md"]

Available Tools / Resources

REST API endpoints for summarization and querying (see next section)
CLI tool for local summarization and batch exports
Web UI for interactive summary inspection and manual exports
Sample prompt templates and summary schemas included in repo
Export adapters: local file, S3, or webhook POST
Logging and metrics endpoints (Prometheus-compatible) for monitoring

API endpoints (examples)

Endpoint	Method	Purpose
/api/v1/summarize	POST	Generate a summary for a provided chat transcript
/api/v1/query	POST	Ask a question about a stored or provided conversation
/api/v1/export	POST	Export a generated summary to a target (S3, webhook, file)
/.well-known/mcp	GET	MCP metadata discovery endpoint

Example POST to summarize a chat:

curl -X POST "http://localhost:8080/api/v1/summarize" \
  -H "Authorization: Bearer <token>" \
  -H "Content-Type: application/json" \
  -d '{
    "conversation_id": "meeting-2026-04-01",
    "messages": [
      {"role": "user", "content": "We need to reduce latency in the API."},
      {"role": "assistant", "content": "What part of the stack is slow?"}
    ],
    "length": "short",
    "format": "md"
  }'

Response (abridged):

{
  "conversation_id": "meeting-2026-04-01",
  "summary": "- Goal: reduce API latency\n- Action: profile service, optimize DB queries\n- Owners: backend team",
  "format": "md"
}

Use Cases

Meeting notes: Automatically produce concise meeting summaries with action items and owners, then export to your team’s wiki or task tracker.
Customer support: Summarize long support interactions into issue descriptions and next steps for engineers or billing teams.
Developer chat analysis: Condense technical discussion threads to capture design decisions, accepted proposals, and unresolved concerns.
Compliance & audit: Create exportable conversation records for audits with configurable redaction and retention policies.
Quick query: Feed a long transcript and ask targeted questions (“What decisions were made about X?”) to get precise answers without re-reading.

Getting Started Tips

Tune prompt templates for your domain: change instructions to emphasize action items, decisions, or technical detail as needed.
Use caching for repeated transcripts to reduce LLM costs.
Start with the Docker image for easiest setup; switch to a custom model provider in config.yml if needed.
Secure the server with API keys or OAuth if deploying to production; treat conversation data as sensitive.

For full details, examples, and configuration options, see the GitHub repository: https://github.com/mcpso/mcp-server-chatsum

ChatSum: MCP Server for LLM Chat Summaries