Jasper
by Jasper
Read a Jasper workspace's brand voices, audiences, style guides and knowledge base, and generate marketing copy through Jasper's own AI engine. Seven tools, all read-only: nothing publishes, and nothing edits a brand voice. OAuth sign-in with one generic scope.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect Jasper via MCP
https://mcp.jasper.ai/Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Jasper Tools & Capabilities (7)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- The single OAuth scope is a wildcard. Jasper's RFC 9728 descriptor at
https://mcp.jasper.ai/.well-known/oauth-protected-resourcedeclares exactly one entry inscopes_supported:mcp. It names no object and no verb, so it cannot exclude anything. Granting it grants whatever the server implements now and whatever Jasper registers later — including, were the update and delete operations its REST API already offers ever to be added, those. The read-only boundary described on this page is enforced by Jasper's choice of tools, - No cost visibility, and no published rate. Two tools spend credits and none can read the workspace's balance, even though Jasper's REST API exposes a usage endpoint. The per-action rate card is the only path Jasper disallows in its
robots.txt, so the cost of a call cannot be looked up before you run one. Do not blanket auto-approvegenerate-contentorrun-jasper-agent. - Publishing exists in Jasper, just not here. Jasper's GEO Agent can publish to a connected Webflow site from inside Jasper's own application, behind an explicit approval step. Whether invoking that agent through
run-jasper-agentcan reach the same path is unresolved — we did not test it, since triggering a publish is precisely the irreversible action this research does not perform. - The read-only annotations understate what a call does. All seven tools carry
readOnlyHint: trueand nodestructiveHint: true, including the two that generate. A client treating that flag as "safe to auto-run" will auto-run billable calls. openWorldHint: falseon a tool with a web-search parameter.run-jasper-agentacceptsWEB_SEARCHin itstoolsarray while declaring it does not interact with the open world. The fetching happens inside Jasper, but the annotation does not convey that.- Business plan required. Jasper restricts API access — and therefore the MCP server — to its Business tier, so this connector is unavailable on lower plans regardless of Claude tier.
- Nothing in the tool surface publishes. If you wanted an agent that drafts and posts, this is only the first half. The connector produces copy; distribution is manual, or belongs to Jasper's in-app Webflow integration, or to a separate automation product such as Zapier.
- Model provenance is unavailable and unselectable. Jasper routes generation across model providers with documented fallbacks, its generation endpoints take no model parameter, and their responses return no model field.
- The deployed remote server was not directly read. We read Jasper's published npm package, whose manifest builds a remote server from the same registration code but ships only the local build. The endpoint requires authentication, so no live
tools/listwas performed and no tool schema was read from the wire. Directory, documentation and source agree on all seven names. - No prompts or resources are claimed for the remote server. The npm package registers three MCP *resources* —
jasper://brandvoices,jasper://audiencesandjasper://styles— alongside the seven tools, all GET-only reads of the same data. Whether the hosted server declares a resources capability could not be checked without authenticating, so treat the resource surface as unverified for the remote endpoint. - The no-training commitment is narrower than it first reads. Jasper's trust page says company data and outputs are "never used to train third-party LLMs" — a claim scoped to third parties, made on a marketing page rather than in the privacy policy, which does not discuss knowledge bases or training at all.
- Rate limits are per workspace, not per user. Jasper's documented defaults are 105 requests per minute on generation and knowledge-search routes, and 200 per minute on the listing routes, with style guides lower at 60 per minute. One busy agent consumes the whole workspace's budget.
Frequently asked questions
No. All seven tools return generated text to the agent and stop there. Jasper's source registers four GET calls and three POST calls, and every POST is a generation or search endpoint. Nothing writes to a CMS, a social account, a campaign or a scheduler, and no tool carries a publish parameter. Publishing stays a copy-paste step you perform yourself.
No, and this is the important one for a marketing team. The three brand tools are plain GET requests against the voices, audiences and styles endpoints. Jasper's REST API does offer update and delete operations for voices and knowledge items, but the connector registers none of them, so a session cannot alter what future generations inherit.
Yes. Jasper's help centre states that API and MCP usage is charged on the credit system, and credits are consumed per action. The per-action rate is not on the open web: the credits rate card is the single path disallowed in Jasper's robots.txt. No connector tool reads usage either, so an agent can spend with nothing able to check the balance first.
Jasper does not commit to one. Its documentation states the AI Engine configures a priority-order of fallbacks per endpoint and per use case to route around third-party model provider downtime. So the model behind generate-content and run-jasper-agent can change between calls, and the connector surfaces no field naming which model produced a given output.
Because Jasper applies the annotation by workspace effect, not by HTTP method. All seven tools declare readOnlyHint true and none declares destructiveHint true, yet three issue POST requests. The label is defensible in that nothing is stored or altered, but it hides that two of those calls spend metered generation quota.
Your entire brand knowledge layer. The tools return every brand voice with its full text, every audience with its description, every style guide including custom instructions and term substitutions, and semantic search across your uploaded knowledge documents with matching text chunks returned inline. That is proprietary company material entering the agent's context.
One scope named mcp, which grants everything the server implements. Jasper's own RFC 9728 descriptor declares scopes_supported as that single value. It names no object and no verb, so consent cannot separate reading a brand voice from spending generation quota, and it would cover any tool Jasper adds later.
A Jasper Business plan and a paid Claude plan. Jasper states that API access is only available to customers on the Business plan, and that the Claude side needs a Pro, Max, Team or Enterprise account. Your network must also reach both mcp.jasper.ai and api.jasper.ai. Generating an API key additionally requires an Admin or Developer role.
Sources
- Jasper MCP Server documentation (retrieved 2026-08-23). Markdown at the same path plus
.md;robots.txtallows this path and namesllms.txt. · retrieved 2026-08-23 - Jasper documentation index,
llms.txt(retrieved 2026-08-23).llms-full.txtreturns 404 on this host. · retrieved 2026-08-23 - Jasper quickstart, Claude (Web) (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper quickstart, Claude (Desktop) (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper quickstart, ChatGPT (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper API rate limits, including the AI Engine fallback note (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper API, Retrieve API usage (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper API, Using Agents (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper API, Getting Started and the Business-plan gate (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper API, Authentication (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper API reference, voice and knowledge lifecycle operations — · · (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper MCP server source,
@gojasper/mcp-serverversion 1.3.0 published 2026-07-23 — package metadata and tarball read from the npm registry athttps://registry.npmjs.org/@gojasper/mcp-server(retrieved 2026-08-23); the human-facing page returns 403 to a plain client. Tool registration, HTTP methods, input schemas and annotations were read from that package. · retrieved 2026-07-23 - Jasper credits-based pricing, stating that API and MCP usage is charged on the credit system (retrieved 2026-08-23). Jasper's help centre sits behind a Cloudflare interstitial that returns 403 to a plain command-line client;
robots.txtpermits the path and it renders normally in a browser. · retrieved 2026-08-23 - Jasper pricing, credits consumed per action (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper
robots.txt, whose only disallowed path is the credits rate card, and which carriesContent-Signal: ai-train=yes, search=yes, ai-input=yes(retrieved 2026-08-23). We did not fetch the disallowed rate card. · retrieved 2026-08-23 - Jasper LLM optimisation page, on automatic model routing (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper sub-processors, naming Anthropic, Cohere, Google Gemini and OpenAI as LLM providers (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper trust page, on data and training (retrieved 2026-08-23) · retrieved 2026-08-23
- Jasper Webflow integration, on GEO Agent publishing with explicit approval (retrieved 2026-08-23) · retrieved 2026-08-23
- Live OAuth posture check: anonymous
initializeto the endpoint returned HTTP 401 withWWW-Authenticate: Bearer realm="MCP Server"; RFC 9728 metadata athttps://mcp.jasper.ai/.well-known/oauth-protected-resource; authorization server metadata athttps://mcp.jasper.ai/.well-known/oauth-authorization-server(2026-08-23) · retrieved 2026-08-23 - Anthropic Connectors Directory entry — , read from our directory snapshot dated 2026-08-23 · retrieved 2026-08-23
- Jasper support — <mailto:support@jasper.ai> · Privacy
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Sales & CRM
- Developer
- Jasper
- Tools
- 7
- Domain
- mcp.jasper.ai
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.