Hermes MCP Gateway
Provides integration with the Hermes agent platform, exposing a curated allowlist of Hermes tools such as web search/extract, browser automation, vision analysis, and skill lookup, along with Hermes/Honcho memory operations, while excluding shell, filesystem, and task-management tools.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Hermes MCP Gatewaysearch the web for MCP gateway security best practices and summarize"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Hermes MCP Gateway
Single localhost-only MCP endpoint that aggregates Context Mode, a strict allowlist from Hermes native tools, one cwd-scoped Samchon Graph code-intelligence tool, three guarded Hermes capabilities, and Hermes/Honcho memory without patching any upstream source.
Endpoint
MCP:
http://127.0.0.1:3060/mcpProcess health:
http://127.0.0.1:3060/healthzFinal-cutover readiness:
http://127.0.0.1:3060/readyz
healthz stays available when a running upstream later degrades. readyz is 200 only when Context Mode, the curated Hermes tools including web, the guarded Hermes capability runtime, and Honcho are all available. Tool discovery is frozen for the process lifetime; provider/config changes require a gateway restart, keeping MCP schemas stable within a running process.
Related MCP server: OpenAI-Compatible MCP Gateway
Tool policy
Context Mode forwards discovered ctx_* tools and remains the primary compact execution/indexing backend. The gateway also exposes a deliberately small native Hermes file/terminal surface for direct host work.
Hermes is default-deny and may expose only:
web_search,web_extractvision_analyzeskills_list,skill_viewnative filesystem:
read_file,write_file,patch,search_filesnative terminal/process:
terminal,processmedia:
image_generate,video_analyzeCamofox-compatible native browser tools:
browser_navigate,browser_click,browser_type,browser_press,browser_snapshot,browser_scroll,browser_back,browser_get_images,browser_console,browser_vision
The gateway does not expose browser_exec, browser_cdp, computer_use, raw Hermes memory/todo state, TTS, or Kanban tools. session_search, delegate_task, and cronjob are exposed only through gateway-owned guarded adapters, not through the stateless Hermes allowlist.
Samchon Graph is exposed as one inspect_code_graph tool. The caller supplies an absolute cwd; the gateway resolves it inside the allowlisted work/source roots and reuses a bounded resident MCP session per project root. The upstream MCP surface must contain exactly that one graph tool.
Guarded Hermes capabilities are:
session_search— read-only search over the local Hermes session database. Cross-profile override is intentionally unavailable; results are historical conversation context, not proof of current external state.delegate_task— synchronous leaf delegation only, one or two children maximum, andconfirmed=trueafter explicit user approval because it spends model inference. Children inherit only Hermesweb,vision,skills, and themcp-context-modetoolset. Context Mode MCP tools are reached through Hermes' scopedtool_search/tool_describe/tool_callbridge; native Hermes terminal/file/code tools and recursive delegation remain out of scope.cronjob— read-onlylist, plus guardedcreate,update,pause,resume,remove, andrun. Mutations requireconfirmed=true. Model/provider/base-URL overrides and script/no-agent/monitor execution fields are not exposed; delivery defaults tolocal.
With the currently pinned Context Mode surface, the public MCP contract is 44 tools: 11 Context Mode + 23 curated Hermes tools (including 10 Camofox browser tools, native file/terminal/process, image generation, and video analysis) + 1 Samchon Graph inspect_code_graph tool + 3 guarded Hermes capabilities + 5 memory + startup_context.
Gateway-owned startup tool:
startup_context— reads only the five fixed Hermes startup files (.hermes.md,SOUL.md,MEMORY.md,USER.md, and canonical PonytailSKILL.md). It accepts no path and is not a generic filesystem surface.
Memory tools are:
memory_profile(read-only)memory_searchmemory_contextmemory_reasoningmemory_conclude
Memory identity is resolved through Hermes Honcho configuration. The existing workspace=hermes and user peer are preserved; only this adapter's assistant peer is chatgpt. Conclusion writes require a durable kind (preference, decision, architecture, project_state) and reject secret-like or explicitly temporary content. List/delete operations retain Honcho semantics.
Browser
The unified ChatGPT gateway exposes the 10 Hermes native browser tools that are available in the pinned Camofox mode. The gateway subprocess uses the same HERMES_HOME=/home/hermes/.hermes and inherits CAMOFOX_URL=http://127.0.0.1:9377, so ChatGPT and Hermes Agent share the same Hermes-managed persistent Camofox userId/profile. Tool calls still use Hermes/Camofox task-session semantics rather than a second browser profile. browser_exec and browser_cdp remain absent because the pinned Camofox backend does not provide those Hermes backends.
Web prerequisite
Hermes natively supports web_search and web_extract through its managed Nous Tool Gateway using the Firecrawl provider. The gateway does not fabricate these schemas and does not implement a second web stack.
The narrow upstream-supported activation path is:
Authenticate the existing Hermes profile with
hermes auth add nous --type oauth --no-browser. This adds Nous OAuth state without requiring the default inference provider to switch to Nous.Select the managed web backend in Hermes config:
web: backend: firecrawl use_gateway: true
Hermes may advertise the web_search/web_extract schemas as soon as the managed backend is selected, even before Nous authentication is usable. The gateway therefore performs one real web_search and one real web_extract smoke call at startup and treats nested provider errors as not-ready. /readyz becomes 200 only when those calls succeed; otherwise the process stays degraded and tunnel cutover is forbidden. Provider/auth changes require a gateway restart so readiness and the MCP schema remain stable for the process lifetime.
Tests
PYTHONPATH=src:/srv/agents/src/hermes-agent \
/home/hermes/.hermes/venvs/hermes/bin/python -m unittest discover -s tests -vLocal integration gates before tunnel cutover:
Context Mode
ctx_executesmoke.vision_analyze,skills_list, native file/terminal/process tools,image_generate,video_analyze, and the exact 10 Camofox-compatiblebrowser_*tools; explicit absence ofcomputer_use,browser_exec, andbrowser_cdp.Guarded capability checks: real
session_search, read-onlycronjob list, and scoped delegation-tool policy; one small delegated child smoke when resource headroom permits.Honcho profile/context/search and controlled
memory_concludecreate/readback/delete.Gateway MCP initialize + exact 44-tool
tools/list+ representative calls, including a read-only Camofox browser navigation/snapshot smoke and cwd-scoped Samchon Graph smoke.healthzandreadyzreadback, including Samchon Graph readiness.systemd restart and enabled-state readback.
Only after
readyz=200: point Tunnel Clientmainfrom3050/mcpto3060/mcp; rollback is the inverse URL change plus Tunnel Client restart.
systemd
The repository ships systemd/hermes-mcp-gateway.service. It uses Wants/After rather than hard Requires, binds the application itself to loopback, imports the same Hermes runtime environment files, and keeps upstream failures from cascading through systemd dependency teardown.
This server cannot be deployed
Maintenance
Related MCP Connectors
Cross-tool persistent memory and context for AI assistants over MCP.
Private-by-default, local-first memory/context/task orchestrator for MCP apps and agents.
AI Reasoning Cache & Consensus Layer with 11 MCP tools via Streamable HTTP.
Model Context Protocol server for the Apideck Unified API. Connect any MCP-compatible agent framework to 100+ accounting systems, HRIS platforms, file storage providers, and more through one integration. More information https://www.apideck.com/mcp-server
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceA single MCP server that gives Claude, ChatGPT, Codex, and any MCP-compatible AI access to the same personal context via two tools: context_get and context_log.1MIT
- FlicenseNot gradedqualityDmaintenanceLocal MCP server that exposes fixed tools for GPT, Claude, and Gemini while routing to any OpenAI-compatible chat completions backend with independent configuration per target.1-
- FlicenseNot gradedqualityBmaintenanceAggregates multiple MCP servers and custom Python tools behind a single endpoint, with intelligent context and tool discovery for AI agents.-
- AlicenseNot gradedqualityAmaintenanceProvides a local context-memory layer for AI assistants, enabling retrieval-augmented queries, explanations, feedback, and status checks via MCP tools.1Apache 2.0