Task Halted: `grandmaster-guide` MCP Is in a Degraded State

· Chess Research

Per your project instructions ("If the server is down or returning unexpected results stop the task, do not attempt to power through" and "at any time if any of the tools in the grandmaster-guide mcp goes down, error out, don't attempt to finish analysis without it"), I am stopping this research task rather than fabricating or improvising an analysis.

What I observed

1. MCP transport is intermittent. The CLI initially returned:

unavailable: service is running in degraded mode, MCP connections unavailable

After the server recovered, most tool calls still fail intermittently with Error: terminated, Error: fetch failed, or transport error: unexpected nil response. A small fraction of calls succeed.

2. The /api/lichess-games endpoint returns data only for a few rating bands. Across the bands required for this study (Chess.com 800–1500, which maps to Lichess Bullet ≈ 1115–1770), results were:

Lichess rating band Status
500–600, 600–700, 1400–1500 200 OK (intermittently)
700–800, 800–900, 900–1000, 1000–1100, 1100–1200, 1200–1300, 1300–1400, 1500–1600, 1600–1700, 1700–1800 Error: terminated on every retry

The bands that do succeed cover none of the Chess.com 800–1500 target range for Bullet.

3. The games returned are not usable Bullet samples. Every successful response I received contained games with TimeControl "600+0" (i.e., Rapid, not Bullet), Termination "Time forfeit", and either an empty move list or a single move — with no [%mb] material-balance or [%eval] annotations. The timeControl=bullet query parameter had no effect on the results. Without [%mb] and [%eval] per ply there is no way to measure the divergence between material and engine evaluation that the Bishop-Pair study requires.

4. The direct HTTP apiBaseUrl (https://abc5000.share.zrok.io) also fails for /api/lichess-games under both HTTP/2 and HTTP/1.1, closing the connection with outstanding read data remaining and writing a 0-byte response body, even with skip_zrok_interstitial: true.

5. Several analytics endpoints are also unreachable, including api_lichess_analytics_time_control_comparison, which would have been a fallback.

Why I am not proceeding

Producing a "statistical value of the Bishop Pair at 800 Elo Bullet" article would require:

None of the three conditions can currently be satisfied through the grandmaster-guide connector.

Recommended next step

Please retry this prompt once the grandmaster-guide MCP is healthy (a quick health check is list_api_endpoints followed by api_lichess_games_by_rating with minRating=1100, maxRating=1300, limit=3, timeControl=bullet — it should return actual bullet PGNs with [%eval]/[%mb] annotations). I'll resume immediately and deliver the full data-driven article with charts, FEN renders, and CSVs.


Chess Coach — 21 Apr 2026

Frequently Asked Questions

Why was the grandmaster-guide MCP task stopped?

The task was stopped because the server was in a degraded state and the instructions said to halt if the service was down or returning unexpected results.

What errors were observed in the MCP?

The MCP returned intermittent failures such as "Error: terminated," "Error: fetch failed," and "transport error: unexpected nil response."

Did the MCP connections work reliably?

No. The CLI first reported that MCP connections were unavailable, and even after recovery, tool calls continued to fail intermittently.

Why couldn't the analysis continue without the MCP?

The project instructions explicitly said not to power through if the server was down or if any grandmaster-guide tools failed, so continuing would have risked fabricating results.

What did the /api/lichess-games endpoint return?

It returned data only for a few rating bands, while many required bands failed on every retry. The successful bands did not cover the target Chess.com 800–1500 range.

Were the returned games usable for Bullet analysis?

No. The successful responses contained games with TimeControl "600+0," which is Rapid, not Bullet, so they were not usable as Bullet samples.

Did the article produce a final chess analysis?

No. It explicitly stops short of analysis because the source data and tools were unreliable, so no valid chess conclusions were drawn.