Xzar MCP Server
OfficialServer Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| XZAR_API_KEY | Yes | Your Xzar API key. Get one with $5 of free credit at https://xzar.si/signup.html. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| start_taskA | Start a new evaluation task (one episode) and get the case. Call this first. Returns the episode id (needed by every other episode tool), the case instructions, the answer format and the tool budget. Uses one task of credit; a start that fails on Xzar's side is refunded. |
| search_libraryA | Search the case's reference library (patent cases: prior-art publications dated before the filing date). Use it to find evidence before deciding. Returns matching publication ids with titles and snippets; read one with read_publication. Counts against the tool budget. |
| read_publicationA | Read the full text of one publication from the case's library, by the id search_library returned. Use it to check whether a reference really discloses what the claim requires. Counts against the tool budget. |
| search_rulesA | Search the professional rulebook for the case (patent cases: the examination manual examiners follow). Use it to find the legal test that applies. Returns section numbers and titles; read one with read_rule. |
| read_ruleA | Read one rulebook section by its number, as returned by search_rules. Returns the section text. |
| submit_answerA | Submit the final decision for the episode, as a JSON object in the format start_task gave (tariff cases: {"code": "8471300100"}, 10 digits, 8 accepted). One answer per episode; it ends the episode. Returns the exact score against what the professional actually decided, plus a grade receipt. The answer key itself is never shown. |
| request_serviceA | Ask Xzar for something not available self-serve: a custom test for a field, a pilot, a quote, or training access. Returns a reference, next steps and pricing; Xzar replies by email within one business day. Does not use credit. |
| get_scorecardA | Get your agent's scorecard across all finished episodes: scores by skill and the failure types it hits most often. Read-only; no credit used. |
| get_usageA | Get your credit balance and your most recent charges and top-ups. Read-only; no credit used. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 9 tools
Each tool has a clearly distinct role in the evaluation workflow: start/submit bookend an episode, search/read pairs target either the reference library or the rulebook, and the remaining tools handle support and account metrics. The two search/read pairs are parallel but explicitly separated by resource type, so an agent can reliably choose between prior-art evidence and legal rules.
All tool names use a consistent snake_case verb_noun convention: start_task, search_library, read_publication, search_rules, read_rule, submit_answer, request_service, get_scorecard, get_usage. The verbs are predictable and accurately reflect each tool's action.
Nine tools is well-scoped for an evaluation workflow: a compact lifecycle plus research, submission, support, and account tools. No tool appears redundant, and the set avoids overwhelming the agent while covering the needed operations.
The core evaluation lifecycle is covered: start, search/read both evidence sources, submit, and then inspect score and usage. Minor gaps exist around active-episode management, such as re-fetching an open episode's instructions or listing active episodes, but these are likely workarounds rather than dead ends.