Atlas / MCP servers / happy520ai / unified-ai-system

unified-ai-systemBLOCK

Name: unified-ai-system
Author: happy520ai

mcp/happy520ai/unified-ai-system

MCP server by happy520ai · a2a protocol · admission control · agent governance · Audited 2026-10-10

Self-hosted model routing gateway and agent control plane for OpenAI, Anthropic, Gemini and A2A agents behind one OpenAI-compatible API, protocol-first. MCP server security, governed tools, prompt enhancement, virtual-key budgets, cache, observability, audit chain and agent governance. Zero-key loca

Verdict

BLOCK

Grade

Trust score

38 /100

Exposed tools

83 65r · 18w · 0d

Transport

stdio · streamable-http

License

Apache-2.0

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

Open-source AI gateway for deterministic prompt enhancement, governed execution, and reproducible verification.

English | zh-CN | Project Site

Unified AI System turns a rough request into a structured, reviewable prompt before execution. It gives

Read from source at commit a44fb804e4bbOBSERVED · 2026-10-10

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code (oci)

claude mcp add mcp-server:0.8.0 --env AI_GATEWAY_MCP_AUTH_TOKEN=${AI_GATEWAY_MCP_AUTH_TOKEN} -- docker run -i --rm ghcr.io/happy520ai/unified-ai-system/mcp-server:0.8.0:None

Exposed tools (83)

65 read · 18 write · 0 destructive.

Tool	Risk	Description
`Documentation`	read	Plan user docs, manuals, or API docs with structure, examples, and acceptance criteria.
`LocalRunner`	read	在当前进程内执行任务(用于开发/测试环境)
`SubProcessRunner`	read	在独立子进程中执行任务(用于生产环境,提供进程级隔离)
`a`	read	A
`add-function`	write	Add a new exported function to an existing module. The function must be
`add-test`	write	Write comprehensive tests for an existing function or module that
`agent_governance_describe`	read
`agent_governance_list`	read	List only the governed Agents visible to the Gateway-authenticated tenant. The tool accepts no tenant or owner override.
`agent_governance_status`	read	Inspect Agent Governance status through the authenticated Gateway identity. The tool accepts no tenant override and fails closed when that identity is not authorized for platform status.
`alpha`	read	a
`ast_edit`	write	AST-aware code editor that understands scope. Supports: rename_identifier (scope-safe rename avoiding strings/comments), extract_function (extract lines into a function), wrap_in_condition, wrap_in_try. Safer than string replacement for identifiers.
`audit_events`	read	Immutable audit log
`b`	read	B
`beta`	read	b
`c`	read	C
`code_format`	read	Format code in a file. Supports JavaScript/TypeScript with basic formatting: consistent indentation (2 spaces), trailing newline, remove trailing whitespace, normalize line endings. Does NOT require external tools like prettier.
`code_run`	write	在显式配置的隔离容器中执行 JavaScript。需要 code:run 权限;未配置隔离后端时拒绝执行。无网络、无项目文件挂载,不在宿主进程运行代码。
`custom_attested_read`	read	attested read
`custom_unattested_read`	read	unattested read
`custom_undeclared_write`	write	undeclared write
`custom_without_permission`	read	missing permission
`d`	read	e
`documented_tool`	read	Reads one record by id. Use
`explain-code`	read	Produce a human-readable explanation of what a piece of code does.
`external_denied_result`	read	Test denial signaling
`external_missing_context`	read	Test missing irreversible context
`external_test`	read	Test irreversible effect
`file_insert`	write	Insert new content at a specific line number in a file. Useful for adding new functions, imports, or sections.
`file_read`	read	Read a file
`file_read_probe`	read	probe
`file_write`	write	写入或覆盖文件内容。需要 file:write 权限。会自动创建父目录。内容限制 50MB。
`fix-bug`	read	Fix a known bug described in a comment or failing test. The fix must
`fixture_write`	write	Write only the owned fixture
`gateway-response`	read	Fake-provider output from Unified AI System.
`gateway_chat`	read
`gateway_health`	read	Inspect gateway health, provider mode, and the real-provider safety flag.
`gateway_prompt_enhance`	read
`gateway_prompt_enhance_llm`	read
`gateway_readiness`	write	Inspect first-run readiness for chat and the local gateway runtime.
`generate_test`	read	Generate a basic test scaffold for a JavaScript/TypeScript source file. Analyzes the file for exported functions and creates a test file using node:test. The generated test imports the source and creates one test case per exported function.
`get_weather`	read	Get weather
`git_branch`	write	List, create, or switch branches. Use for branch management operations.
`git_commit`	write	Stage files and create a commit. Always review changes with git_diff before committing.
`git_create_pr`	write	Create a Pull Request using the GitHub CLI (gh). Requires gh to be installed and authenticated. The current branch will be used as the head branch.
`git_diff`	read	Show differences between working tree and index, between commits, or between branches. Use this to review changes before committing.
`git_log`	write	Show commit history. Returns structured commit information including hash, author, date, and message.
`git_push`	write	Push commits to a remote repository. This is a dangerous operation that requires explicit approval. Always verify with git_status and git_log before pushing.
`git_status`	read	Show the working tree status. Returns modified, staged, untracked, and branch information.
`glob`	read	Fast file pattern matching tool. Returns matching file paths sorted by modification time. Supports glob patterns like
`governed_budget_probe`	read	governed sanitizer budget probe
`governed_effect_probe`	read	committed effect audit probe
`governed_key_probe`	read	governed key sanitizer probe
`governed_output_probe`	read	governed output probe
`grep`	read	A powerful content search tool using regular expressions. Searches file contents and returns matching lines with optional context. Use this to find code patterns, function usages, imports, etc.
`image_read`	read	Read an image file and return it as visual context. The image will be included in the response for visual analysis.
`knowledge_readiness`	read	Inspect the gateway knowledge infrastructure without loading or changing data.
`knowledge_retrieve`	read
`lookup`	read	Look up the fixture
`lsp_definition`	read	查找符号定义(跳转到定义)。给定文件路径和行列位置,返回符号的定义位置。
`lsp_hover`	read	获取符号类型信息和文档。返回 hover 信息,包括类型签名、JSDoc/docstring 等。
`lsp_references`	read	查找符号引用。返回符号在项目中被使用的所有位置。
`lsp_symbols`	read	列出文件中的所有符号(函数、类、变量等)。返回文档符号的层次结构。
`ops__create_ticket`	write	Create a ticket
`read_file`	read	Read one file
`recorded_search`	read	test
`refactor`	read	Refactor existing code to use a specified pattern (e.g. replace
`s`	read	d
`semantic_search`	read	Semantic code search using TF-IDF scoring over code files. Finds code chunks most relevant to a natural language query. Use for
`shell_exec`	write	执行 shell 命令。需要 shell:exec 权限。命令会被分类标记风险等级。
`strat`	read	desc
`test`	read	x
`test_tool`	read	A test
`type_check`	write	Run TypeScript type checking on a file or project. Uses
`web_fetch`	read	获取 URL 的网页内容并提取文本。需要 network:fetch 权限。
`web_search`	read	Search the web for up-to-date information, documentation, API references, tutorials, and more. Returns titles, URLs, and snippets.
`workflow_actions`	read	List workflow action definitions without invoking any action.
`workflow_executions`	read	Execution records
`workflow_health`	read	Inspect the governed workflow subsystem without starting a workflow.
`workflow_run`	write
`workflows`	read	Workflow definitions
`workforce-analysis`	read	Workforce role-based analysis from Unified AI System.
`workforce_agents`	read	List configured workforce agent descriptors without dispatching them.
`workforce_health`	read	Inspect the workforce subsystem without planning or executing work.

Trust audit

BLOCKgrade F · trust 38/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	PASS
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	PASS
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (10 observation(s))
Network: declared (16 observation(s))
Shell: declared (6 observation(s))
Dependencies: not all pinned
Secrets in source: found

Findings (25)

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

apps/ai-gateway-service/src/capabilities/neuronCodeGenerator.js:92

{ pattern: /eval\s*\(/, label: "禁止 eval()" },

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

apps/ai-gateway-service/src/capabilities/safetyScanner.js:16

message: "Code contains eval() which is a security risk",

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

packages/forge-core/src/multi-agent-review/builtin-rules.js:42

name: 'eval() usage',

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

packages/forge-core/src/multi-agent-review/builtin-rules.js:46

message: 'eval() usage is dangerous and may allow code injection.',

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

packages/forge-core/src/multi-agent-review/builtin-rules.js:327

name: 'eval()/exec() usage',

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHFilesystem / path · fs.credential_store · CWE-22, CWE-59

apps/agent-console/src/agentTaskCommands.ts:46

".netrc", ".git-credentials", "credentials", "credentials.json", "auth.json", "evidence"]);