Tokenaissance 远程 MCP 服务器是什么?
Tokenaissance 的远程 MCP 服务器是一个托管的 Model Context Protocol 端点,地址是 https://tokenaissance.com/mcp,AI 助手无需在本地安装或运行任何东西即可连接。
Tokenaissance MCP Server已验证已精选
@Rain
基本信息
关于 Tokenaissance MCP Server
Tokenaissance is a online Personal Agent for Quant and Research investment available at https://tokenaissance.com.
连接信息
https://tokenaissance.com/mcp接入方式
claude mcp add tokenaissance --transport http https://tokenaissance.com/mcpTokenaissance MCP Server 工具
23List the agents whose skills you can read over this connection, with the name and id to pass to list_skills and read_skill.
List the skills one of your agents publishes, with what each one is for. Use read_skill to load one.
Read one file of one published skill, named as list_skills lists it. Defaults to the skill's SKILL.md, which is the file that explains how to perform the skill.
Append one instruction to a task that already exists (start_task creates one), and answer as soon as the server has accepted it — the reply is the task's occupancy at that moment, never the turn's text, and the connection is not held waiting. Read the answer later with read_task.
Take back the OLDEST instruction still waiting in that task's queue, so it never runs. Answers {withdrawn:true, turnId} naming which submission went — with several clients queued, "the earliest" and "mine" are different answers. Withdrawing something already running is stop_task.
Stop the turn that is RUNNING in that task. Cooperative: answers {stopped:false} when nothing was running, and never promises what the task does next — read_task says that. Taking back a QUEUED instruction is withdraw_task.
List the tasks one of your agents has, newest first, with what each one is about. Timestamps live here and only here — the other tools answer about what is happening, not when.
Start a NEW task for one of your agents and give it the first instruction. Answers with the task handle, and with that task's resource URI (`task://<agent>/<task>`), as soon as the server has accepted it — read the reply later with read_task. To be told when the task changes state: on 2026-07-28 send `subscriptions/listen` with that URI in `notifications.resource_subscriptions`; on the legacy handshake there is no push channel, so poll `read_task`. What that stream carries is the RAW activity — the text as it is written and the text the turn had written when it was stopped — while read_task answers with the processed result; use the stream for progress, and read_task for the answer. Use run_task with a `task` to add work to one that already exists.
Fold one more instruction into the turn that is RUNNING in that task, so the agent reads it at its next step instead of starting a separate turn. Answers {steered:false} when nothing is running — use run_task then, which queues behind the current turn.
Read a task: what you may do next, and how its last completed instruction ended. {status:"idle"|"running"|"queued"} says whether the task can take work now — while it is running or queued no outcome is reported, because with no handle to point with the server cannot say whose completion it would be. Once nothing is in flight, {outcome:"replied", reply} carries the answer text; {outcome:"empty"} says it ended without text, {outcome:"stopped"} that it was stopped before it finished, and {outcome:"failed"} that a model/provider failure ended it. A "waiting" count says how many instructions are queued behind the holder. Pass full:true for the whole archived message projection. This is also how a legacy-handshake client follows a task at all: there is no push channel on 2025-06-18 / 2025-11-25, so poll it with backoff; on 2026-07-28 use subscriptions/listen and keep read_task for the answer. This is the PROCESSED view and the authority for the answer: runtime-injected messages are skipped, long results are clipped, and a stopped or failed turn reports no reply even if it had written text. The subscription stream for the same task carries the raw activity instead — more detail, including fragments this read does not count as an answer.
Create a new agent in your account (like "New agent" in the web app). It starts with no skills; install them with install_skill, then drive it with run_task.
Install a skill from a registry into one of your agents, using the same shape the web app's installer sends: a registry plus a skill name, or a repository.
Read the account's spendable credits and whether it is blocked. Credits are bought in the browser (the pricing page); this tool only reports the number a client needs before submitting work.
Change one task's title. Answers {renamed, cardRefreshed}. When the task is shared, its curation card is re-derived from the new title; `cardRefreshed:false` means the rename stands and the card still shows the old title, so read the card again before you report it changed.
Delete one task and withdraw its share first, so no public link can outlive the conversation. Answers {deleted, shareRevoked}; `shareRevoked:true` means a live share was withdrawn by this call, and `false` means there was none. A task that is already gone answers success, because the caller asked for it to be absent.
Share one of your tasks and answer {token, url, isActive}. This is the "add" of this surface: it creates the share, or restores a withdrawn one on the SAME token, so an old link works again. Anyone with the url can read the conversation.
Answer whether one of your tasks is shared, and with which token, url, board state and `filePath`. A task that was never shared answers {shared:false} — that is an answer, not an error. This tool never creates a share.
Set, change, or clear the default file a visitor opens from the share url. `filePath` is required, and `null` clears it — an absent parameter cannot be told from a deliberate clear. The task must already be shared; share_create is the way to create one.
Withdraw the share of one of your tasks, so the url stops working and any board listing drops it. Answers {revoked:true} when a live share was withdrawn and {revoked:false} when there was nothing to withdraw. The token is kept, so share_create restores the same url.
Post an EXISTING share of one of your tasks to the public curation board, which also allows search engines to index it. Requires its own authority: publishing is a public act, so it is not folded into sharing. The task must already be shared; call share_create first. Answers {curated:true, token, url}.
Remove an existing share from the public curation board and stop feeding it to search engines. The share itself stays, so the url keeps working and curation_publish can list it again. A page a search engine copied may stay in its index until it crawls again.
List your own shares, newest first, with each token, url, task, board state and filePath. Add `agent` to narrow the list to one agent. The view count is not here: it costs one query per share, and share_status answers it for one share.
Read the public curation board: the shares their owners posted for everyone. Answers the cards, newest and featured first, plus the total count. This is public content, so it needs the public read authority rather than the account one. An empty board answers an empty list.
常见问题
怎么连接 Tokenaissance 的 MCP 服务器?
把端点 https://tokenaissance.com/mcp 添加到任意支持 MCP 的客户端,比如 Claude Code、Cursor 或 VS Code。本页的接入代码块可以一步配置好每个客户端。
Tokenaissance 的 MCP 服务器需要认证吗?
需要。Tokenaissance 使用 OAuth——第一次连接时,你的 MCP 客户端会打开浏览器窗口登录并授权,之后会复用这个凭证。
Tokenaissance 的 MCP 服务器用的是哪种传输方式?
Tokenaissance 暴露的是 Streamable HTTP 端点,这是远程 MCP 服务器普遍使用、主流客户端都支持的传输方式。
更多远程 MCP 服务器
1
1cent
Production remote MCP server with 35 safe web-intelligence tools: 32 paid x402 URL operations plus 3 free discovery/demo tools. Base Mainnet USDC, no account or API key; SSRF protection, strict schemas and idempotent set
A
AccountHub
One MCP connection for all your Gmail, Google Calendar, Drive and Contacts accounts, Slack workspaces and Notion — free.
3
3GPP Scout
Hosted 3GPP MCP server: cited search across Rel-15 through Rel-20 TS/TR specs, recent TDocs, company proposals and meeting decisions. OAuth 2.1; no local index required.
AWS Marketplace
Discover, evaluate, and buy solutions for the cloud
AI Video MCP by AITuber
AI Video MCP by AITuber is a hosted MCP server that lets AI agents create and publish video. You connect it once, then ask your agent for a video in plain language. It writes the script, picks a voice, generates the visu
评论