Skip to content
mcp/skillhub

Capxule MCP Server

by capxuleai.capxule/capxulev2026.10.13

The context marketplace: expert guided journeys your assistant runs step by step.

0remoteofficial registry

context tax

~5,752 tok/turn12 tools

security

92/100

cold start

hosted

freshness

Active5d ago

Our reading of Capxule, from the data

Capxule is an MCP server from capxule, listed in the official MCP Registry.

Its 12 tools add about 5,752 tokens to every model turn (measured 2026-10-19).

Maintenance is active: the last public signal was 5 days ago (2026-10-14).

It runs as a hosted endpoint, so there is nothing to install.

Install Capxule MCP server

Install in Claude Code

claude mcp add --transport http capxule-ai https://capxule.ai/mcp

Remote endpoints

  • streamable-httphttps://capxule.ai/mcp

Context tax

Connecting this server adds ~5,752 tokens of tool definitions to every model turn (heavy, 12 tools). Measured 2026-10-19 via its hosted endpoint; o200k tokenizer, worst case for clients that load all tools up front. How we measure

ModelPer 1,000 turnsWith prompt caching
Claude Opus 5.5$23.01$1.15
Claude Sonnet 5.5$11.50$1.15
Claude Haiku 5.5$0.575—

Input prices as of 2026-10-06. Excludes your own prompt and output.

Tools (12)

  • turn847 tok

    Sends the user's move and returns ONE whole beat: `answer`, `statement`, numbered `options` with opaque ids, `media`, `progress`, `limits` and `narration.text_fallback` (ready to relay). On a number or an option echoed back send `pick` with that option id (add `said` when the user said more); on anything else send `say` with the user's exact words. `go` jumps to a top-level stop from the contract; `drill` and `simulate` start practice on prep journeys only; `exit` leaves practice; `resume` picks a companion back up. Costs one turn (a simulation start costs two). Supply a fresh `turn_id` per st

  • account_bind786 tok

    Starts binding an account for the user without a web visit. Ask for an email in the user's own words, read back the masked address, then call with method email_code (the default); the user reads you the six-digit code and you call account_verify. Alternatives: magic_link (a link they tap), sms_code (a number the user named, only where SMS is available; otherwise the reply says so), device (a CLI agent: the user types a short code in a browser; poll account_verify with the returned challenge), resume_code (a code from another assistant's handoff). Costs nothing and creates nothing until the cod

  • find_capsule610 tok

    Covers: 163 guided journeys across 17 areas: Finance (33, e.g. Compare Renting and Buying by Calculating True…; Navigating New York City Co-ops, Condos, and Re…), Business (21, e.g. Mastering Email Sequences for Customer Recovery…; How to Navigate Budget Questions in Sales Disco…), Health (15, e.g. Managing Sleepless Nights Through Action, Rules…; How to Dispute Medical Bills Effectively), Education (12, e.g. Win More Financial Aid Support Before Enrollment; An Educator's Guide to Spaced and Retrieval Pra…), Lifestyle (12, e.g. Choosing Childcare: Balancing Cost, Quality, an…; Home Solar Decis

  • account_verify526 tok

    Completes account_bind: pass the challenge it returned and the code the user read you (digits only; spaces are fine). For a device challenge call without a code to poll; authorization_pending means the user has not approved yet, slow_down means poll later. Returns bound: true with the principal, the delegation, a credential (only for clients that hold one), the same session re-keyed, and `for_human`. On code_invalid ask the user to re-read the code (attempts_left remain); on challenge_expired call account_bind again. Costs nothing. Rules (binding): Relay `answer` and every `options[].text` exa

  • open_journey487 tok

    Opens (or resumes) a guided journey and returns the session contract: `session` (keep it), the stop list (titles only), the capabilities, the limits in turns, the rights, and the first beat (static, costs nothing). Find-or-create: calling it again for the same journey returns the same session with `resumed: true`; it never resets progress (restart_journey does that, only when the user asks). Pass `goal` when the user stated one in their own words. On the open mount a stable `_meta.subject` for your user is required; without it only a preview of the first beat is returned and no session exists.

  • make_artifact345 tok

    Composes the user's takeaway (kind "takeaway") from THIS conversation: what they asked, what landed, the stops they visited. Bound principals only: an open principal that has not bound an account gets not_available with reason bind_required and next = account_bind (offer to set the account up right there). Returns markdown plus rights and a cite line; deliver the markdown as-is. Costs 2 turns; one per session (a repeat returns the same artifact free). Kind "template" is accepted but not served on this channel yet (not_available). Rules (binding): Relay `answer` and every `options[].text` exact

  • browse_cataloghttp335 tok

    Pages through the published guided journeys, optionally within one category (a slug from list_categories), featured first. Returns the same cards as find_capsule plus total and next_offset. Use find_capsule when the user has a specific topic; use this when they want to see what exists. Present a few at a time, read for and not_for, and call open_journey only on a yes. Free; never costs a turn. Never say a price. Rules (binding): Relay `answer` and every `options[].text` exactly as given. You may put only `statement` in your own words. On a number or an option echoed back, send `pick` with that

  • journey_state317 tok

    Where the user is in the journey: position (stop, visited of total), `remaining` (stops left, percent done, and each top-level topic's visited of total), mode, the goal line, the chosen way, practice stats, limits and a `resume_line` you can say. Hosts that render apps show it as a map card. Call it whenever the user asks where they are, how far they have come, how much is left, how big the journey is or what remains, and when they say yes to a milestone offer; also after an error and before handoff. Costs nothing; may be cached ttl_ms. Rules (binding): Relay `answer` and every `options[].text

  • restart_journey297 tok

    Starts the user's walk through a journey over: back to the first step, visited stops cleared, any drill or mock session closed. Turns already used stay used and the practice record is kept. Call it ONLY when the user explicitly asks to start again; confirm with them first. Returns the same session contract as open_journey. Rules (binding): Relay `answer` and every `options[].text` exactly as given. You may put only `statement` in your own words. On a number or an option echoed back, send `pick` with that option id. On anything else, send `say` with the user's exact words, unsummarised. Never a

  • handoff241 tok

    Mints a resume link and code for this session so the user can continue on the website or in another assistant. Call it before switching hosts or ending; give the user `resume_url` (or read them `resume_code`) and say `resume_line`. Costs nothing. Rules (binding): Relay `answer` and every `options[].text` exactly as given. You may put only `statement` in your own words. On a number or an option echoed back, send `pick` with that option id. On anything else, send `say` with the user's exact words, unsummarised. Never ask for, accept or relay a password. If the user types one, do not send it anyw

  • account_status140 tok

    Whether this connection is bound to an account, whether that account includes the premium journeys (`premium`, a flag) and unlimited web turns, the assistants connected to it (`delegations`, with their scopes in plain words), `manage_url` for the human (a web page; never a checkout), and for a machine principal its live pass. Costs nothing. Never say a plan name or a price to the user; say what they can do.

  • list_categories75 tok

    Lists the areas the published guided journeys cover: each category slug, a readable label, how many journeys it holds and a few example titles. Use it when the user asks what is available, or before browse_catalog. Free; never costs a turn. Never say a price.

Security signals

92/100 — no known issues found as of 2026-10-19, from the checks below. This is not a guarantee of safety; review what a server can do before connecting it.

Deductions

  • No license detected-5
  • No security policy (SECURITY.md)-3

Checks

  • ✓Repository hygiene (archived, activity, license, security policy)
  • ✓Tool descriptions (prompt-injection and hidden-text heuristics)
  • ○Package metadata (install scripts, provenance) — pending
  • ○Source scan (secrets, CVEs, risky code) by vectra-guard — pending

Scan it yourself before installing with vectra-guard, the open-source agent security CLI. Scoring rules

Freshness

Active — last maintenance signal 5d ago. The newest of the signals below sets the band.

  1. Last commit (default branch)

    2026-10-14 · 5d ago · GitHub

  2. Latest release

    no data · GitHub

  3. Package published

    no data · npm/PyPI

  4. Registry entry updated

    2026-10-14 · 5d ago · official registry · v2026.10.13

FAQ

›How do I install the Capxule MCP server in Claude Code?

Run: claude mcp add --transport http capxule-ai https://capxule.ai/mcp. For Cursor, VS Code, Claude Desktop and Windsurf, use the install tabs above.

›Does Capxule require an API key?

No required environment variables are declared in its published metadata.

›Can I use Capxule as a remote (hosted) MCP server?

Yes. It is a hosted MCP server; connect to its URL with any client that supports remote MCP.

›Is Capxule in the official MCP registry?

Yes, as ai.capxule/capxule.

›How many tokens does Capxule add to my context?

Its tool definitions total about 5,752 tokens, resent on every turn.

Sources and method

Official MCP Registry entry
updated 2026-10-14
GitHub repository
fetched 2026-10-19
Last checked against all sources
2026-10-19

The description, packages and links come from the maintainer's own published metadata. Token cost, security signals, freshness and the reading above are MCPSkillHub's own measurements; see the methodology. Spot an error? Request a correction or claim the listing.