SkillAgentSearch skills...

boj-server

One MCP stdio endpoint for a whole toolchain — GitHub, GitLab, cloud, mail, browser and research tools — with capability-gated dispatch, a machine-checked ABI, and cartridges fetched on demand from boj-server-cartridges. Zero runtime dependencies.

Install / Use

claude mcp add hyperpolymath -- npx -y github:hyperpolymath/boj-server

If the server publishes to npm under a different name, use that package instead — check the repo README.

About this skill
🔌

MCP Server

Model Context Protocol server

Quality Score

83/100

Category

Security

Supported Platforms

Claude Code
Claude Desktop

Our assessment of boj-server

boj-server scores 83/100 on our quality scale, 383rd of 544 Security skills we index.

Its MCP Server is 20 KB long, well organised into 20 sections with 6 code examples: a thorough specification that gives an agent plenty to work with.

It has 3 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
30/30
Structure
20/20
Description
15/15
Adoption
3/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated 4 days ago, so boj-server is actively maintained.
  • Our last check on 2026-09-10 found the source still online.
  • It is released under MPL-2.0, a copyleft license: you can use it, but modified versions you distribute must carry the same license.
  • Its trust signals score 92/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

Safety scan

No issues found

Our scan of the first 100 KB of the file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands.

Automated pattern scan on 2026-09-26. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.

boj-server compared with similar skills

All 4 of these similar skills score higher than boj-server; compare them before choosing.

SkillScoreStarsUpdatedFormat
boj-server (this skill)by hyperpolymath8334d agoMCP Server
Agent-Reachby Panniantong10085.4k10d agoCLAUDE.md
headroomby headroomlabs-ai10073.8ktodayCLAUDE.md
rufloby ruvnet10073.3k1d agoCLAUDE.md
CowAgentby zhayujie10047.1ktodayCLAUDE.md

Frequently asked questions

How do I install boj-server?
Run claude mcp add hyperpolymath -- npx -y github:hyperpolymath/boj-server. The install tabs above show the steps for each supported agent.
Which AI agents does boj-server work with?
It is written for Claude Code and Claude Desktop, as a MCP Server file. Other agents that read the same format can often use it too.
Is boj-server safe to use?
Our scan of the first 100 KB of the file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. It is MPL-2.0-licensed and scores 92/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is boj-server still maintained?
The repository was last updated 4 days ago, so boj-server is actively maintained.
<!-- SPDX-License-Identifier: CC-BY-SA-4.0 --> <!-- GENERATED from README.adoc by standards/.github/workflows/readme-derive-reusable.yml — do not edit. -->

One MCP endpoint for the WHOLE toolchain — GitHub, GitLab, Cloudflare, Vercel, Verpex, Gmail, Calendar, browser automation, research, ML, multi-agent coordination, and a large catalogue of pluggable domain cartridges, all reachable through a single zero-dependency stdio bridge.

License: MPL-2.0 npm Glama MCP Server OpenSSF Best Practices OpenSSF Scorecard Software Heritage Quality gate

What it is, honestly: BoJ exposes 68 MCP tools today (45 boj_* + 23 coord_*) over stdio with zero runtime dependencies. It catalogues 125 domain cartridges, but most of those are an inspectable catalogue, not live services — a cartridge only performs real actions when its backend process is running and you supply the right credentials. The bridge is fully inspectable offline; side-effectful tools return a structured {error, hint} until their backend is up. See Cartridges for the full story.

Contents

Features

  • Unified endpoint — GitHub/GitLab, Cloudflare/Vercel/Verpex, Gmail/Calendar, Firefox browser automation, CodeSeeker code intelligence, Semantic Scholar research, and Hugging Face ML, all behind one MCP server.

  • 68 MCP tools — 45 boj_* (5 core discovery/dispatch + explicit high-frequency tools) and 23 coord_* multi-agent coordination tools.

  • 125-cartridge catalogue — a single boj_cartridge_invoke reaches any catalogued cartridge; explicit boj_<domain>_<verb> tools exist for the highest-frequency operations.

  • Multi-instance AI coordination — local-coord-mcp lets several Claude / Gemini / Codex sessions on one machine discover each other, claim tasks without collision, and run under a master/journeyman/apprentice supervision model.

  • Zero runtime dependencies — the bridge runs on Node, Deno, or Bun with no install step.

  • Inspectable offline — boj_health, boj_menu, boj_cartridges, and boj_cartridge_info answer from an offline manifest so clients can introspect the server without any backend running.

  • MCP resources & prompts — 7 boj:// resources and reusable prompts (audit-repo, convene-cluster, deploy-with-dns-ssl, summarize-channel, triage-issues, proof-status).

  • Hardened — per-call rate limiting, size caps, prompt-injection detection with Unicode-confusable normalisation, and error sanitisation (paths, stack traces, and env vars stripped from responses).

  • Formally verified core — the coordination ABI is written in Idris2 with discharged proof obligations; remaining axioms are documented, not hidden.

Install

BoJ ships as an MCP server over stdio. The published npm package (@hyperpolymath/boj-server) has zero runtime dependencies, so no install step is ever required regardless of runtime.

Most cartridges call the BoJ REST backend on http://localhost:7700. Without it, the server is still fully inspectable; side-effectful tools return {error, hint}. See Backend.

Claude Code (CLI)

claude mcp add boj-server -- npx -y @hyperpolymath/boj-server@latest

Claude Desktop

Edit claude_desktop_config.json:

  • macOS — ~/Library/Application Support/Claude/claude_desktop_config.json

  • Windows — %APPDATA%\Claude\claude_desktop_config.json

  • Linux — ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "boj-server": {
      "command": "npx",
      "args": ["-y", "@hyperpolymath/boj-server@latest"],
      "env": { "BOJ_URL": "http://localhost:7700" }
    }
  }
}

Restart Claude Desktop after saving.

npx (any MCP client)

The minimum stdio spec is command: npx, args: ["-y", "@hyperpolymath/boj-server@latest"]. Optional env: BOJ_URL (default http://localhost:7700). This works with VS Code / Copilot, Cursor, Cline, Windsurf, Continue.dev, Zed, and the Gemini CLI — point each client’s MCP config at that command. This repo’s .mcp.json is a working reference config.

Deno / Bun / Node (from a clone)

The bridge entrypoint is mcp-bridge/main.js and runs on any of the three runtimes with no install:

# Deno (no install step; the project's documented runtime)
deno run -A /path/to/boj-server/mcp-bridge/main.js

# Bun (zero-install)
bun /path/to/boj-server/mcp-bridge/main.js

# Node (>= 18)
node /path/to/boj-server/mcp-bridge/main.js

Quickstart

After install, ask your LLM: "Use the boj_health tool." You get {status:"ok", uptime_s, version} when the backend is up, or a structured hint when it is offline.

To talk to the bridge directly over stdio, send newline-delimited JSON-RPC. Initialize, then list tools:

printf '%s\n%s\n' \
  '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"demo","version":"0"}}}' \
  '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}' \
  | node mcp-bridge/main.js

The initialize response reports protocol 2024-11-05 and server boj-server; tools/list returns 68 tool definitions (45 boj_*, 23 coord_*), each carrying a full description, JSON-Schema inputSchema/outputSchema, and MCP behaviour annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint).

Call a tool:

{ "jsonrpc": "2.0", "id": 3, "method": "tools/call",
  "params": { "name": "boj_health", "arguments": {} } }

The server also implements resources/list (7 boj:// resources) and prompts/list.

Capabilities overview

The bridge exposes 45 boj_* tools and 23 coord_* tools. A subset of cartridges have explicit boj_<domain>_<verb> tools for high-frequency operations; everything catalogued is reachable through boj_cartridge_invoke.

| Group | Tools | Examples | |-----------------------------------------|-------|------------------------------------------------------------------------------------------------------------------------------------------------------------------| | Core discovery / dispatch | 5 | boj_health, boj_menu, boj_cartridges, boj_cartridge_info, boj_cartridge_invoke | | GitHub | 14 | boj_github_list_repos, boj_github_create_issue, boj_github_create_pr, boj_github_merge_pr, boj_github_search_code, boj_github_graphql | | GitLab | 8 | boj_gitlab_list_projects, boj_gitlab_create_mr, boj_gitlab_list_pipelines, boj_gitlab_setup_mirror | | Browser (Firefox) | 7 | boj_browser_navigate, boj_browser_click, boj_browser_type, boj_browser_read_page, boj_browser_screenshot, boj_browser_tabs, boj_browser_execute_js | | Cloud | 3 | boj_cloud_cloudflare, boj_cloud_vercel, boj_cloud_verpex | | Communications | 2 | boj_comms_gmail, boj_comms_calendar | | Research / code intel / ML / search | 4 | boj_research, boj_codeseeker, boj_ml_huggingface, boj_search | | Coordination (local-coord-mcp) | 23 | coord_register, coord_claim_task, coord_send, coord_review, coord_approve, coord_health |

Set BOJ_TOOL_SCOPE=core to advertise only the discovery surface; explicit boj_<domain>_* tools remain reachable via boj_cartridge_invoke regardless. A CSV of prefixes (e.g. core,github,browser) advertises core plus named groups.

Multi-agent coordination (coord_*)

A localhost multi-agent bus (default 127.0.0.1:7745) lets multiple AI sessions on one machine discover each other, claim tasks without collision, and operate under supervision (master approves; journeyman executes; apprentice stays gated):

  • Peers — coord_register, coord_list_peers, coord_set_variant, coord_set_capabilities, coord_get_peer_capabilities.

  • Typed envelopes — coord_send, coord_send_gated, coord_receive (Nickel-contract validation, opt-in strict mode).

  • Task claims — coord_claim_task with role-based watchdog TTL, coord_progress heartbeats, coord_sweep_watchdog, optional advisory paths for path_overlap warnings.

  • Track record — coord_report_outcome, coord_get_affinities, coord_set_declared_affinities, coord_scan_suggestions (emits overclaim/drift advisory envelopes).

  • Supervision — coord_review, coord_review_entry, coord_approve, coord_reject, coord_promote_to_master, coord_transfer_master, plus coord_status / coord_health.

Task-claim collision-freedom is a task-level guarantee, not a git-level lock: two journeymen claiming different tasks that touch the same file can still hit a vanilla merge conflict. The supported pattern is branch-per-claim + per-peer worktree, advisory path-claims, and master-gated integration. The companion terminal UI lives in coord-tui/ and at hyperpolymath/coord-tui.

Cartridges

BoJ catalogues 125 cartridges across trust tiers (Teranga / Shield / Ayo). Be clear about what that means:

  • Catalogued ≠ live. boj_menu lists the full catalogue, but most cartridges report available: false. They are entries describing a capability — its API base URL, auth model (often brokered through vault-mcp), and any native FFI path — not a running service.

  • A cartridge becomes available when (1) its backend process is running and reachable via the BoJ REST API, and (2) you have supplied the credentials it needs.

  • Credentials are typically environment variables (GITHUB_TOKEN, GITLAB_TOKEN, CF_API_TOKEN, OAuth tokens, …) or are brokered by the vault-mcp credential cartridge. boj_cartridge_info <name> returns the cartridge’s manifest, including the exact auth requirement.

Truncated for display — read the full file on GitHub.

Related Skills

View on GitHub
GitHub Stars3
CategorySecurity
Updated4d ago
Forks0

Languages

JavaScript

Trust signals

92/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 low