dsa-coach-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| DSA_COACH_DB | No | Where revision cards are stored | ~/.dsa-coach/coach.db |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
Coach DSA practice from the user's own history on competitive-programming platforms. Call list_platforms first to learn each platform's tags and levels, get_profile to see where the user stands, and search_problems with filters you chose from that data. When the user tells you their username on a platform, save it with link_account; tools that take a handle then default to it, and the dashboard shows that account's stats. After the user solves or revisits a problem, record how well they recalled it with log_review (again, hard, good or easy). On a platform with an adapter, call review_evidence with their handle first: it reads the rating from their submissions, which is more reliable than their own estimate. Log every mistake they made too: label each mistake with a short kind, reusing a label from mistake_kinds whenever one fits. revisions_due lists what to revise now, least likely to be remembered first; mistake_patterns and list_mistakes show where the user keeps going wrong. For a platform with no adapter, read the problem page (for example with a browser MCP server) and call record_problem with its URL, title and tags before log_review.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| list_platformsB | Connected platforms, with the tags and levels each one accepts as filters. |
| get_profileB | The user's standing on one platform: level, solved and unsolved counts per tag, and their most recent submissions. Without a handle, uses the one linked with link_account. |
| review_evidenceA | Suggest a log_review rating from the user's submissions to this problem since it was last reviewed: again if none was accepted, good if the first was, hard if rejected ones came first. Never easy, since speed is not in the history. Where the platform hides rejected submissions, or nothing was submitted, it suggests nothing and says why; ask the user then. Tell the user the suggestion and its reason, and let them overrule it. Without a handle, uses the one linked with link_account. |
| link_accountB | Save the user's username on a platform, replacing any saved before, after checking it exists. Tools that take a handle default to it, and show_dashboard shows this account's stats. Returns those stats. |
| unlink_accountB | Forget the username saved for a platform. |
| search_problemsB | Problems having every tag given, at the level given. Take tags and levels from list_platforms or from problems the platform returned. With exclude_solved_by, skips that handle's solved problems. |
| log_reviewA | Record how well the user recalled a problem, schedule the next revision with FSRS, and journal any mistakes made on the way. |
| revisions_dueC | Problems due for revision now, the one the user is least likely to recall first. |
| record_problemA | Record a problem from a platform this server has no adapter for, as read from its page: its URL, title, the tags the page shows and its difficulty as the page states it. The platform is the URL's host, and the problem id its path. Returns the platform and problem_id to pass to log_review. |
| progressB | Per platform the user has reviewed on: problems reviewed, reviews logged, and problems due for revision now. |
| schedule_infoB | The FSRS parameters in use, whether they were fitted to this user's own reviews, and how many reviews are logged. |
| personalize_scheduleB | Fit the FSRS parameters to this user's own review history, so revisions are timed to how fast they personally forget. Needs the optimizer extra. |
| mistake_kindsA | Every mistake label used so far, most frequent first. Reuse these labels when logging new mistakes so the same kind of mistake is counted together. |
| list_mistakesB | Journaled mistakes, newest first, with the problem each was made on. Filter by
label, problem tag, platform, one problem (platform and problem_id), or the last
|
| mistake_patternsB | Mistake counts per label, broken down by problem tag, platform and month. Counts only; reading what they say about the user is up to you. |
| show_dashboardB | Show the user's progress per platform, the revisions due now ordered by how likely they are to be forgotten, a heatmap of mistakes by label and tag, and what the history says about their memory: recall per tag now and two weeks out, which mistakes stopped recurring, problems rated easy then failed, and how well FSRS's predictions matched, plus live stats for each account linked with link_account. Hosts that support MCP Apps render it as an interactive dashboard. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| practice_plan | Plan the next practice session from the user's own mistakes and progress. |
| weekly_report | Report on the user's last week of practice and say what to do next. |
| revision_quiz | Quiz the user on today's revisions using the mistakes they made before. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| dashboard | Interactive view for show_dashboard. |
TDQS
Scored across 16 tools
Most tools have distinct jobs, but several report overlapping stats: get_profile (per-platform standing), progress (per-platform reviewed/due counts), and show_dashboard (which also includes per-platform progress and 'revisions due now') largely restate each other at different verbosity levels, and revisions_due duplicates part of show_dashboard. The three mistake tools (mistake_kinds, list_mistakes, mistake_patterns) are better separated but still near each other. Descriptions help, but an agent could plausibly pick progress vs show_dashboard vs get_profile arbitrarily.
Roughly half the tools use verb_noun (list_platforms, search_problems, log_review, link_account, record_problem), but the rest are bare nouns (progress, revisions_due, schedule_info, mistake_kinds, mistake_patterns), including one verb_noun-ish outlier (review_evidence). The convention is readable but clearly mixed rather than predictable.
16 tools is slightly heavy but each maps to a real concern in this domain: platform discovery, account linking, problem search/recording, review logging, FSRS scheduling, and mistake analytics. No tool looks redundant enough to be cut outright, though progress could arguably fold into show_dashboard.
The lifecycle is broadly covered: discover platforms, link/unlink accounts, find or manually record problems, log reviews with FSRS rescheduling, inspect schedules, and analyze mistakes. The notable gap is that review and mistake records are create-only — there is no way to edit, delete, or undo a mis-logged review or mistake, and no corrections path beyond re-logging.