M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
H

HyperChat

Community
Not a standard server
A local AI agent platform where configuration is code and tools come in through MCP

This is not a typical connect-and-use MCP server; read the source repository before using it.

How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #362 of 438
Stars
★ 709 Very popular
Runtime
Node.js
Credentials
API key / credential required
License
Other / unspecified
Last commit
⚠
39FMRS · D

HyperChat is a local-first AI chat and agent platform that organizes agents, memory, and MCP services around a configuration-as-code model, with two entry points: a Web multi-workspace mode and a CLI agent-first mode. The README is detailed about environment-variable layering, agent configuration, and custom commands, but says nothing about specific MCP tools, transports, or client setup, and the source contains no server manifest (not cached). The license is marked NOASSERTION in the source and the README only points to a LICENSE file, so confirm the terms before use.

Strongest · Documentation 9/20 Weakest · Security and permissions 6/20

Reliability
7/20
Security and permissions
6/20
Maintenance
9/20
Documentation
9/20
Setup experience
8/20
Why each score
Reliability 7/20
The README states HyperChat 1.0 was hand-written and is migrating to 2.0, an inconsistent version narrative, and no server manifest is available (SERVER MANIFEST: [not cached]), so declared tools cannot be checked against real behavior or against a reproducible MCP init + tool-list handshake. No source files, no committed tests, and the build.yml badge is illustrative rather than verifiable test evidence; under static calibration reliability must not exceed 12. Runnable entry points are merely inferred from the npm package name and CLI examples, so a happy path is plausible, but error handling, dependency control and edge cases are unsupported; deducted to 7.
Security and permissions 6/20
The README explicitly grants Agents filesystem, terminal, browser, git and database MCP tools, and sample agent configs set isConfirmCallTool: false, i.e. tool calls execute without confirmation against the local filesystem and shell, so least privilege and confirmation of dangerous operations are unclear. API keys and the web password are injected via env vars and .env, and examples use placeholders such as sk-1234567890, so no real token leak, malware, credential theft or covert exfiltration is evident. However, no reviewed code shows permission checks, path scoping, network-egress disclosure or data-flow boundaries, so it cannot score high; rated 6, and no 0-4 red line is triggered.
Maintenance 9/20
The repo is not archived, has 707 stars and 30 open issues, and the README declares active 2.0 work with community 'Vibe Coding', while the npm package and CI badge indicate a release/build pipeline, showing some activity. But the license reports as NOASSERTION and the README only says the project uses an open-source license with details in the LICENSE file, which is not verifiable here; there is no clear ownership, versioning policy, security-response channel or dependency-update evidence, and issue responsiveness is unproven. Governance and licensing gaps cap this at 9.
Documentation 9/20
Documentation is fairly layered: a dual-architecture comparison table, a five-level environment variable priority scheme, agent.yaml and Markdown custom-command templates, multi-@ file reference examples, and Web/CLI usage scenarios are all presented with readable snippets, making install and configuration largely usable. Yet there is no MCP server manifest or tool-parameter-level documentation, no disclosure of permission boundaries, cost/quota or troubleshooting guidance, and the README contradicts itself on 1.0/2.0 capabilities (license, vision), matching the 'usable but hidden assumptions and troubleshooting gaps' anchor near 10; rated 9.
Setup experience 8/20
Setup offers npm install -g @dadigua/hyperchat or npx, and env vars can be set five ways (export, global/workspace .env, CLI flags) with clear examples, a short and plausible path. However, no configuration example for any mainstream MCP client (e.g. Claude Desktop, Cline) is provided, nor a platform compatibility matrix or required Node runtime version, and no package config is available in the reviewed root to confirm the CLI entry point actually exists. Per static calibration setup cannot exceed 15 absent real CI workflows plus committed tests covering key paths; evidence here is weaker still, so 8.

Static review · not runListed 2026-09-10

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesRuns commands or codeUses the networkControls a browser

Best for

  • Developers who want AI configuration stored locally and managed in Git
  • Users who need to reuse a personal agent across projects while keeping per-project context
  • Users who want agents to act on local resources such as the file system through MCP
  • Engineering teams that need a command-line entry point for automation

Not for

  • Users who just want a hosted, zero-configuration cloud chat service
  • Users who require official commercial support or SLAs from the project maintainers
  • Users looking for a ready-made MCP tool list and client setup config, which the README does not provide

Required permissions

  • Local file system read/write: the README states agents can operate on the local file system directly through the MCP protocol
  • Reading the project's .hyperchat/ configuration, agent memory, and chat logs
  • Reading environment variables and .env files (API keys, service port, Web access password)
  • Starting a local Web service that listens on a port (default 16100)
  • Calling external LLM provider API endpoints

Risks and side effects

  • Once an agent is granted file-system or terminal MCP tools, it may modify local files, possibly without confirmation; the isConfirmCallTool setting in agent.yaml affects whether calls are confirmed
  • Configuration, agent memory, and chat logs are stored as plain local files, so committing them to Git or sharing the directory can leak sensitive information
  • Credentials such as API keys live in environment variables or .env files and can be committed by mistake or read by other local processes
  • If HYPERCHAT_WEB_PASSWORD is not set, the local Web service may be reachable by other processes or machines on the same network
  • Custom command templates send the referenced @ file contents to the configured LLM provider
  • The README states the project is for learning and exchange only and describes no security guarantees or audit mechanisms

Setup

Before you start

Runtime：Node.js

HYPERCHAT_WEB_PASSWORD optionalsecret Password for the Web UI; set it yourself.
Other optional settings (6)
HYPERCHAT_PORT optional Web service port, defaults to 16100.
HYPERCHAT_HOST optional Bind address, e.g. localhost.
HYPERCHAT_LANGUAGE optional UI language, zh or en.
HYPERCHAT_LOG_LEVEL optional Log level, e.g. info.
HYPERCHAT_OPENAI_BASE_URL optional Custom OpenAI API endpoint.
HYPERCHAT_CLAUDE_BASE_URL optional Custom Claude API endpoint.

The README offers two paths: 1) Install globally with npm install -g @dadigua/hyperchat, or run it directly with npx -y @dadigua/hyperchat. 2) Develop from source: git clone https://github.com/BigSweetPotatoStudio/HyperChat.git, then npm install, followed by npm install inside packages/electron and packages/web, and finally npm run dev. Before use, configure environment variables such as HyperChat_API_KEY, HyperChat_API_URL, HyperChat_AI_Provider (e.g. openai/claude/gemini/kimi/qwen) and HyperChat_AI_Model (e.g. gpt-4o). Start the Web UI with hyperchat serve (default http://localhost:16100), or invoke an agent directly with hyperchat agent <name> "...".

Check that it works

After a global install, run hyperchat serve and open http://localhost:16100 (port configurable via HYPERCHAT_PORT); seeing the multi-workspace Web UI and getting a chat reply from an Agent confirms the install works.

Troubleshooting

  1. Verify HyperChat_API_KEY, HyperChat_API_URL, HyperChat_AI_Provider, and HyperChat_AI_Model; configuration follows a 5-level priority (defaults < process.env < global .env < workspace .env < CLI arguments), so check whether a higher layer overrides your setting
  2. If the Web UI is unreachable, check HYPERCHAT_PORT (default 16100) and HYPERCHAT_HOST, and whether an access password was configured
  3. If an agent is missing, run hyperchat agent list to see agents available globally and in the workspace
  4. If an agent cannot use an MCP tool, check whether allowMCPs in agent.yaml includes that service and whether the workspace mcp.json service pool defines it
  5. If a custom command does not trigger, confirm it is a .md file under the agent's commands/ directory and invoked with a / prefix
  6. For more detail, use --verbose or adjust HYPERCHAT_LOG_LEVEL

Things to try

Once connected, you can ask your AI assistant things like:

  • Analyze the code quality of @./src/index.ts
  • Compare @./package. and @./yarn.lock
  • Run the /review command on @./src/api/user.js
  • Run: hyperchat agent coder "/bug-fix @./src/login.ts login redirect fails"

Use cases

Defining project-specific agents, memory, and MCP service pools in a project-local .hyperchat/ directory
Using the CLI in agent-first mode for quick conversations or for embedding into scripts and CI/CD flows
Managing several project workspaces, agent sets, and chat histories side by side in the Web multi-workspace UI
Reusing fixed prompt workflows through Markdown-defined slash commands
Analyzing or comparing several files at once by referencing multiple @ file paths

Overview

HyperChat is a local AI agent platform (its Chinese README describes it as pioneering an "AI as Code" idea) that uses APIs from various LLMs for its chat experience and implements productivity tools through the MCP protocol. It emphasizes full localization: data stays on the machine, and configuration lives as files under a .hyperchat/ directory that travels with the project and can be version-controlled with Git. The project uses a two-tier architecture: a Web front end acting as a multi-workspace collaboration hub, and a CLI front end optimized for agent-first interaction. Agents support a memory file (memory.md), MCP services declared in agent.yaml, and Markdown-template-driven custom commands such as /bug-fix and /review. The README notes that HyperChat 1.0 was a fully hand-written project and is being migrated to 2.0. No MCP tool inventory, transport details, or client connection config examples are provided in the source.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 3f9cf0341c44 Data synced 2026-10-19 Read the FMRS scoring method