Agent Mesh

io.github.google由 GitHub 帳號 google 發布v0.2.0更新於 Nov 3, 2026

以你自己的已認證 Agent Mesh 成員身份發現並呼叫遠端 MCP 工具。

AI 翻譯由 AI 模型根據下方的原文描述翻譯,可能有誤。

原文描述

Discover and call remote MCP tools as your own authenticated Agent Mesh member.

發布方自己的服務:AgentMesh

已驗證STDIO僅桌面

概覽

Agent Mesh 的本機 MCP 服務;無需登入;STDIO。

暫無 AI 概覽:以下資訊來自註冊中心條目和服務本身。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Agent Mesh,將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。

其他 MCP 客戶端

參照 儲存庫 中的啟動說明。

README

@agentmesh-p2p/sdk

Native JavaScript SDK for joining an Agent Mesh from inside the agent process. It replaces the agentmesh-node sidecar for agents written for Node.js or running in a browser page: the agent enrolls with the control plane, joins the mesh through a router, finds services and calls them, answers A2A requests for the agent itself, and follows the control plane's keys, bans and policy while it runs. It publishes no service; a tool or a model others should find by name runs behind a agentmesh-node.

Guide: google.github.io/agentmesh/docs/guides/native-sdks, from an empty machine to two programs on a mesh. Source: github.com/google/agentmesh/tree/main/sdk/js.

Install

bash
npm install @agentmesh-p2p/sdk @modelcontextprotocol/sdk zod

Requires Node.js 22.18 or later. In a browser, bundle it with the page (the package's browser field selects the browser files); the guide's In a browser section has the details and an example page.

MCP client over stdio

The package's agentmesh executable joins as its own mesh member and exposes five MCP tools: get_mesh_info, discover_remote_services, find_remote_tools, describe_remote_tool and call_remote_tool. A stdio-capable MCP client starts and stops the process; no node daemon is needed for this caller. Providers still publish their services through agentmesh-node.

To try the executable from a source checkout:

bash
cd sdk/jsnpm cinpm run build

Use absolute paths in your MCP client's configuration:

json
{  "mcpServers": {    "agentmesh": {      "command": "node",      "args": ["/absolute/path/to/agentmesh/sdk/js/dist/cli.js"],      "env": {        "AGENTMESH_CONTROL_PLANE_URL": "https://your-control-plane.example",        "AGENTMESH_BOOTSTRAP_TOKEN_PATH": "/absolute/path/to/join-token",        "AGENTMESH_STATE_DIR": "/absolute/path/to/this-client-identity"      }    }  }}

Get the control plane URL and token file from your mesh operator or agentmesh-one. Alternatively, set AGENTMESH_JWT_PATH to a workload/OIDC token file; do not set both token paths. Credentials are read from files, never command-line values.

The first run enrolls and saves its identity; later runs resume from the state directory without the bootstrap token. JWT enrollment rereads the token file when refreshing. Use a distinct state directory for each client identity and mesh; do not run multiple processes against the same directory. The default is ~/.config/agentmesh/mcp. Plain HTTP is allowed on loopback only unless AGENTMESH_INSECURE_CONTROL_PLANE=true explicitly permits it elsewhere.

Stdout contains only MCP messages; diagnostics go to stderr. Closing stdin or sending SIGINT/SIGTERM shuts down the mesh session. Tool calls use the member's existing authenticated streams and policy, including provider label checks. Remote tool results retain text, non-text content, structured output and errors. Discovery follows tool pagination, rejects looping catalogs and bounds each search to 64 providers, 256 services per provider, 256 tools per service and 1024 results. Requests have a 10-second deadline; narrow a large search by peer or service. intent is accepted but ignored, matching the node's current tool.

This member publishes no services, so it has no list_local_services tool. Inference and A2A discovery return metadata, but this executable provides no HTTP proxy, /v1 endpoint, local socket or local_proxy_url. Use an SDK client or an agentmesh-node HTTP proxy to invoke those protocols.

Use

Both programs below are in examples/ and run against a real mesh in the repository's tests. They read the mesh from AGENTMESH_CONTROL_PLANE_URL and the enrollment token from AGENTMESH_BOOTSTRAP_TOKEN_PATH; the guide shows how to get both from agentmesh-one or from the operator of an existing mesh.

Find a service and call it:

ts
// Calls something on the mesh: a tool of an MCP service or a path of an// inference or A2A service someone published, found by name, or an agent that// published nothing, reached by its peer ID.////   node call.js mcp://everything echo '{"message": "hi"}'//   node call.js inference://ollama /v1/models//   node call.js 12D3KooW... a2a://agent /card//// AGENTMESH_CONTROL_PLANE_URL names the mesh. The first run enrolls with the file// AGENTMESH_BOOTSTRAP_TOKEN_PATH (a token the mesh operator gave you) or// AGENTMESH_JWT_PATH (a workload identity token your platform issues, such as a// Kubernetes projected service account token), and keeps the identity and// credential in AGENTMESH_STATE_DIR; later runs resume from there without it.import { homedir } from "node:os";import { AgentMesh, type DiscoveredProvider } from "@agentmesh-p2p/sdk";
let argv = process.argv.slice(2);// A first argument that is not a service target is the peer ID of an agent.let peerId: string | undefined;if (argv[0] !== undefined && !argv[0].includes("://")) {  [peerId, ...argv] = argv as [string, ...string[]];}const [service = peerId !== undefined ? "a2a://agent" : "mcp://everything", toolOrPath = peerId !== undefined ? "/card" : "echo", args = '{"message": "hi"}'] = argv;
const mesh = await AgentMesh.enroll({  controlPlaneUrl: process.env.AGENTMESH_CONTROL_PLANE_URL ?? "https://mesh.example.com",  bootstrapTokenPath: process.env.AGENTMESH_BOOTSTRAP_TOKEN_PATH,  jwtPath: process.env.AGENTMESH_JWT_PATH,  stateDir: process.env.AGENTMESH_STATE_DIR ?? `${homedir()}/.config/agentmesh/caller`,  // A plaintext http:// control plane is otherwise accepted only on loopback.  allowInsecure: process.env.AGENTMESH_INSECURE_CONTROL_PLANE === "true",});const session = await mesh.join();console.log(`on the mesh as ${session.peerId}`);
// The peer to call: an agent named by ID, or a provider of a published service.let provider: DiscoveredProvider = { peerId: peerId ?? "", addrs: [] };if (peerId !== undefined) {  // An agent is not in the discovery table; the SDK finds the path to its  // peer ID through the routers.  await session.connect(peerId);} else {  const providers = await session.discover(service);  if (providers.length === 0) {    throw new Error(`no member of the mesh serves ${service}`);  }  // A provider record can outlive its member; the first that answers is used.  let reached = false;  for (const candidate of providers) {    try {      await session.connect(candidate);      provider = candidate;      reached = true;      break;    } catch (err) {      console.error(`${candidate.peerId}: ${(err as Error).message}`);    }  }  if (!reached) {    throw new Error(`no provider of ${service} is reachable`);  }}console.log(`${service} is served by ${provider.peerId}`);
if (service.startsWith("mcp://")) {  const tools = await session.listTools(provider, service);  console.log(`tools: ${tools.map((t) => t.name).join(", ")}`);  const result = await session.callTool(provider, service, toolOrPath, JSON.parse(args));  console.log(result.text.join("\n"));} else {  const response = await session.request(provider, service, toolOrPath);  console.log(response.status, response.text());}
await session.close();

Be an agent others can call. Nothing is published; a caller reaches the agent by its peer ID, the one it prints, through a router:

ts
// An agent on the mesh: joins, then answers A2A requests from other members// until stopped. It publishes nothing. There is no service name to look up; a// caller reaches the agent by its peer ID, through a router, as a2a://agent.// The mesh policy decides which members may call; the SDK turns the others// away before anything reaches this code.////   node agent.js                        # answered by the handler below//   node agent.js http://127.0.0.1:9999  # forwarded to an A2A server beside it//// AGENTMESH_CONTROL_PLANE_URL names the mesh. The first run enrolls with the file// AGENTMESH_BOOTSTRAP_TOKEN_PATH (a token the mesh operator gave you) or// AGENTMESH_JWT_PATH (a workload identity token your platform issues, such as a// Kubernetes projected service account token), and keeps the identity and// credential in AGENTMESH_STATE_DIR; later runs resume from there, as the same// peer, without it.import { homedir } from "node:os";import { AgentMesh } from "@agentmesh-p2p/sdk";
const [backendURL] = process.argv.slice(2);
const mesh = await AgentMesh.enroll({  controlPlaneUrl: process.env.AGENTMESH_CONTROL_PLANE_URL ?? "https://mesh.example.com",  bootstrapTokenPath: process.env.AGENTMESH_BOOTSTRAP_TOKEN_PATH,  jwtPath: process.env.AGENTMESH_JWT_PATH,  stateDir: process.env.AGENTMESH_STATE_DIR ?? `${homedir()}/.config/agentmesh/agent`,  // A plaintext http:// control plane is otherwise accepted only on loopback.  allowInsecure: process.env.AGENTMESH_INSECURE_CONTROL_PLANE === "true",});const session = await mesh.join();
// Answers every path with who was asked and who asked; a real agent runs an// A2A server here, or beside this process at backendURL.const target = await session.acceptA2A(  backendURL !== undefined    ? { url: backendURL }    : { handler: (request, caller) => Response.json({ name: "agent", path: new URL(request.url).pathname, caller: caller.peerId }) },);console.log(`accepting ${target} as ${session.peerId}`);
const stop = () => void session.close().then(() => process.exit(0));process.on("SIGINT", stop);process.on("SIGTERM", stop);

enroll reuses the identity and credential saved in stateDir when they are still valid for that control plane, and needs exactly one of bootstrapTokenPath, bootstrapToken, jwtPath or jwt (a token string or a () => string | Promise<string> callback invoked at enrollment and on every refresh) otherwise. Read tokens from a file, a callback or the environment; do not put them on a command line. labels are attested at enrollment; join({ egressRequireLabels }) is the floor every peer the session calls must attest, all of it, held for the session.

A plaintext http:// control plane is accepted only on loopback. Pass allowInsecure: true for a network you trust.

License

Apache-2.0. Issues and contributions at github.com/google/agentmesh.

來源:sdk/js/README.md,提交 1c88bcf

工具

0
SourceWeft 尚未在沙箱中啟動過該軟體包,因此其工具未知。

版本歷史

1
  1. v0.2.0最新Nov 3, 2026

評分

已滿足適用檢查項的 50%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通25 分

    請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。

  • 已宣告認證方式10 分

    請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。

  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分

其他 Agentmesh 整合