Nexus Mcp

io.github.j7anPublished by GitHub account j7anv2.1.3Updated Oct 25, 2026

Delegate tasks to coding agents (Claude, Codex, Antigravity) from any MCP client

VerifiedSTDIODesktop onlyDeveloper ToolsAI & ML

Overview

AI-generated overview

Local MCP server that lets any MCP client delegate coding tasks to Claude Code, Codex or Antigravity agents and get their final answer back.

What you can do
Run a single agent turn on a chosen backend (claude, codex or antigravity) with a prompt and an absolute project directory, returning output, session id and usage data. Continue or fork an existing conversation by session id, pick a model, and set a timeout. Bound each call with permission profiles: read_only, workspace_write or full_access. List available backends as a resource, showing which extras are installed, their models and install hints.
When to use it
Useful when you want an MCP client to hand coding or review work to an external coding agent rather than doing it itself, including read-only review passes or sandboxed edits inside a project directory.
Requirements
Runs locally via uvx or pip; each backend is an optional extra that bundles its CLI (roughly 40–150 MB). Authentication is the agent's own: Claude login or ANTHROPIC_API_KEY, codex login, and GEMINI_API_KEY or Vertex AI variables with ADC for Antigravity. On Windows, Claude Code must be on PATH.
Before you install
The full_access profile lets the agent do anything in the project, and workspace_write allows file edits and sandboxed Bash; denied actions are refused without prompting. Agent credentials and session data live in the agents' own directories. Model calls may incur provider costs.

Installation

In SourceWeft

  1. Open Nexus Mcp in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

nexus-mcp

MCP server that delegates tasks to coding agents — Claude Code via the Claude Agent SDK, Codex via the Codex App Server, and Antigravity via the Antigravity SDK — from any MCP client.

Install

Backends are optional extras; install the ones you use.

ExtraBackendAdds
claudeClaude Agent SDKclaude-agent-sdk (bundles the Claude Code CLI, ~100 MB)
codexCodex App Serveropenai-codex (bundles the Codex CLI, ~120–150 MB)
antigravityAntigravity SDKgoogle-antigravity (bundles the Antigravity Go harness, ~40 MB)
allevery backend
bash
uvx --with 'nexus-mcp[all]' nexus-mcp          # run directlypip install 'nexus-mcp[claude]'                 # or installpip install 'nexus-mcp[codex]'                  # Codex onlypip install 'nexus-mcp[antigravity]'

Claude Code MCP config:

json
{ "mcpServers": { "nexus": { "command": "uvx", "args": ["--with", "nexus-mcp[all]", "nexus-mcp"] } } }

Authentication is the agent's own: log in with claude, or set ANTHROPIC_API_KEY; for Codex, log in with codex login (credentials in ~/.codex are shared). Antigravity uses GEMINI_API_KEY, or Vertex AI with GOOGLE_GENAI_USE_VERTEXAI, GOOGLE_CLOUD_PROJECT, and GOOGLE_CLOUD_LOCATION, authenticated through Application Default Credentials (ADC). It does not use Google-account login.

Windows: recent claude-agent-sdk releases ship no Windows wheel with a bundled CLI; install Claude Code so claude is on PATH.

Tool: prompt

Runs one agent turn and returns its final answer.

ParameterDefaultMeaning
backendrequiredclaude, codex, or antigravity
promptrequiredTask for the agent
cwdrequiredAbsolute project directory
profileread_onlyPermission profile (below)
session_id—Continue this conversation
forkfalseBranch session_id into a new conversation
modelprovider defaultModel id or alias
timeout600Seconds before the turn is cancelled

Returns {backend, session_id, output, usage}. Conversations are stored by the agent itself (~/.claude/projects, $CODEX_HOME/sessions, ~/.nexus-mcp/antigravity), so a session_id keeps working after nexus-mcp restarts. Antigravity does not support fork=true.

Permission profiles

Actions outside the chosen profile are denied automatically — nobody is prompted.

ProfileAllows
read_onlyRead/Glob/Grep; git diff/log/show with --no-ext-diff --no-textconv
workspace_writePlus file edits inside cwd and Bash inside the OS sandbox
full_accessEverything

Codex enforces profiles with its OS sandbox (read-only, workspace-write, danger-full-access) and never asks for approval.

Antigravity enforces workspace containment in its harness and uses the OS sandbox for commands. workspace_write fails if that sandbox is unavailable.

Resource: nexus://backends

JSON list of {name, installed, models, hint}; hint gives the install command for a missing extra. Codex lists its models; Claude and Antigravity report null (Antigravity does not provide a model list).

Configuration

VariableDefaultMeaning
NEXUS_CLAUDE_SETTINGS_PROFILEisolatedClaude settings to load: isolated (none), project (project .claude/), inherit (all, including user settings). Profiles still bound every tool call.

Migrating from v1

v1v2
prompt(cli="codex") (codex exec)prompt(backend="codex", cwd=…) (App Server)
prompt(cli=…, execution_mode="yolo")prompt(backend=…, profile="full_access", cwd=…)
batch_promptParallel prompt calls from the client
agent_start / agent_status / agent_resultprompt (synchronous)
agent_continue / agent_forkprompt(session_id=…) / prompt(session_id=…, fork=true)
agent_reviewprompt(profile="read_only", prompt="Review …")
agent_cancelCancel the tool call in the client
agent_respond, elicitationRemoved — choose a profile instead
Preferences, model tiers, MCP promptsRemoved
OpenCode runners and toolsRemoved
NEXUS_* variablesRemoved except NEXUS_CLAUDE_SETTINGS_PROFILE

Development

bash
uv sync --all-extras --all-groupsuv run pytest                    # unit + e2euv run pytest -m integration     # real CLI (slow, makes model calls)uv run mypy src/nexus_mcp && uv run ruff check . && uv run ruff format --check .

License

MIT

Source: README.md at commit e9204e9

Tools

0
SourceWeft started this package once in an isolated sandbox, without credentials, to list its tools. That failed: the package did not install.

Version history

9
  1. v2.1.3LatestOct 25, 2026
  2. v2.1.2Oct 17, 2026
  3. v2.1.1Oct 3, 2026
  4. v2.1.0Oct 3, 2026
  5. v2.0.2Oct 2, 2026
  6. v2.0.1Sep 26, 2026
  7. v2.0.0Sep 26, 2026
  8. v1.1.0Sep 19, 2026
  9. v1.0.3Sep 16, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts