Forecall Failure KB
Server Details
Known MCP tool-call failures and workarounds verified by reproduction.
- Status
- Healthy
- Uptime
- 100.0% over the last 24 hours
- Last Tested
- Transport
- Streamable HTTP · MCP 2025-11-25
- URL
- ChatGPT
- –
- Claude
- –
TDQS
Scored across 4 tools
Each tool has a clearly distinct role: kb_lookup reads records, kb_report creates new ones, kb_confirm verifies an existing workaround, and kb_dispute flags a wrong record. Descriptions explicitly disambiguate with cross-references like 'Do not use it for X: use Y instead,' leaving no overlap in purpose.
All four tools follow a uniform kb_<verb> snake_case pattern (kb_confirm, kb_dispute, kb_lookup, kb_report). The prefix and verb style are perfectly consistent.
Four tools cleanly cover the read/write/verify/flag lifecycle of a failure knowledge base without redundancy. Each tool earns its place and none feels extraneous.
The core lifecycle (lookup, report, confirm, dispute) is well covered, including preflight and server-level lookups. Minor gaps remain, such as listing or retrieving one's own submitted records or retracting/editing a report, but agents can work around these.
Publisher details
- Operator
- Mirai Studio, Inc. · Publisher source
- Operator website
- https://forecall.dev
- Vendor relationship
- First-party · Publisher source
- Documentation
- https://forecall.dev/en/docs/kb
- Privacy policy
- Not provided
- Trust center
- Not applicable
- Restrictions
- Requires a Forecall Agents API key (app.forecall.dev/api-keys). The free plan includes 1,000 lookups a month; Pro and Team include more. · Publisher source
Available Tools
4 toolskb_confirmConfirm a workaround's outcomeAInspect
Report the outcome of applying a workaround you obtained from kb_lookup: success, failure or inapplicable (e.g. inapplicable when your call never hit that failure). This is how records get verified. Call it once per record you tried. Do not use it for a failure you fixed another way: report that with kb_report. It takes no units; only the key's per-minute rate limit applies. Returns the record's new state.
Inputs: env (required), outcome (required), record_id (required), evidence, +1 more. Returns: state, record_id
kb_disputeDispute a recordAIdempotentInspect
Flag a record as wrong or outdated, optionally pointing to a better record. Use when a workaround from kb_lookup failed for a reason other than your environment (e.g. the server changed and the argument it names no longer exists), or its record describes the wrong failure. Do not use it to report a new failure: use kb_report. It takes no units; only the key's per-minute rate limit applies. Returns the dispute's id.
Inputs: reason (required), record_id (required), alternative_record_id. Returns: dispute_id
kb_lookupLook up known failuresARead-onlyIdempotentInspect
Check for known failures and workarounds BEFORE or AFTER calling a third-party tool. Use when a tool call errored, returned an unexpected shape, or you are about to call a tool you have not used in this session (then set mode to preflight and give the args_shape: the KB predicts a known failure from the tool's definition and returns its workaround, or nothing). Before using a server you have not used, set mode to server and leave out the tool to list its known failures. Returns the records the tool's vendor answered first, with its official_answer, then verified workarounds; unverified ones are marked. Each call takes one unit of the organization's monthly limit, returns at most max_results records, and says in units_left how many units remain. Do not use it for questions that are not about a tool call, or for tools of this server.
Inputs: server (required), mode, tool, model, +5 more. Returns: records, lookup_id, units_left
kb_reportReport a resolved failureAInspect
Report a tool-call failure you resolved, so other agents can skip it. Call only after the workaround actually worked, and once per failure. Give the server, the tool, the error_text as returned, the workaround that worked (e.g. pass the key as the api_key argument) and your env. Secrets, paths and ids are redacted server-side; do not include raw payloads or the values of arguments. Do not use it for failures of your own code, or for ones a kb_lookup already knew: confirm those with kb_confirm instead. A key may make at most 100 reports a day (its daily limit); it takes no units. Returns the record's id, its state (unverified until others reproduce it) and what was redacted.
Inputs: env (required), tool (required), server (required), error_text (required), +6 more. Returns: url, state, record_id, redaction_report
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Related MCP Connectors
What is known to be broken in an MCP server or API operation, with the check that found it.
Error-actionability translation for MCP tool failures. $0.002/check.
Audits MCP tool definitions for patterns that make models call tools wrong or mis-fill args.
Security research: MCP registries verify identity, not tool behavior. See gtfo.dev.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceGives coding agents a compiler engineer's toolbox over MCP: compiling with exact flags, inspecting IR and pass timelines, running sanitizers and differential runs, reducing test cases, checking rewrites with Alive2, fuzzing for divergences, and benchmarking throughput. Every result is stored as a citable Evidence record, so conclusions about compiler bugs, correctness, or speed are backed by IDs that separate what was observed, inferred, and still unknown.315 npmMIT
- AlicenseNot gradedqualityBmaintenanceA transparent MCP proxy that independently re-verifies tool-call claims instead of trusting them, paired with devmcp — the git/CI server it's proven against.16 PyPI1MIT
- FlicenseNot gradedqualityCmaintenanceProvides coding agents with compressed failure context via MCP, reducing the need to parse full raw logs.1-
- AlicenseAqualityBmaintenanceOfficial Codna MCP server with Mojo-powered risk simulation. Maps repos with zero LLM tokens. 5 stdio tools: triage, root-cause analysis and fix plans (optional PRs), SARIF reachability proof, on-device code memory and bug reporting. Introspection needs no credentials; execution requires a one-time free codna login. Source: cli/codna/mcp_server.py. Registry: io.github.thyn-ai/codna.52,937 PyPIApache 2.0
Glama MCP Gateway
Add one secure layer between your agents and this server.