Local-first MCP memory and code intelligence that helps coding agents recall repo context, decisions, failures, and review impact.
Jambavan has 29 tools with mostly complete schemas and descriptions, but significant gaps reduce overall quality. 24 tools have descriptions (83%), but 5 deprecated alias tools lack meaningful descriptions. Input schemas are present for all tools, but many lack parameter-level descriptions or type constraints. Tool naming is generally clear (verb_noun pattern), but composition issues exist: jambavan_memory_* tools (store, read, search, forget, mine_session) could be consolidated; jambavan_failure_* duplicates memory patterns. Error handling is minimal, no recovery guidance or actionable error messages visible. Output schemas are undocumented. STDIO transport caps protocol readiness at 50, limiting overall score despite decent definition work.
Execute a bash command in the project root (requires JAMBAVAN_ALLOW_BASH=1).
Deprecated alias for jambavan_decompose_task (counsel protocol).
Deprecated alias for jambavan_dev_rules (counsel protocol).
Bind the MCP server to a project root directory for indexing and analysis.
Compress prose content while preserving facts and structure using token-aware summarization.
Search the index for symbols and retrieve relevant code context by query.
Report on code quality issues, technical debt, and inline comments marked with 'rin:'.
Five deprecated alias tools (root_cause, verify_gate, strategy_plan, decompose_task, dev_rules) have minimal descriptions ('Deprecated alias for jambavan_*') that do not explain what the tool does or when to use it. LLMs cannot determine tool purpose from 'Deprecated alias' text alone.
Output schemas are not documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., what does jambavan_context return? Does it include file paths, line numbers, symbol types?). This forces agents to guess or make exploratory calls.
Memory and failure tools (jambavan_memory_*, jambavan_failure_*) duplicate functionality and naming patterns. jambavan_memory_store, jambavan_memory_read, jambavan_memory_search, jambavan_memory_forget could be consolidated into a single jambavan_memory tool with action parameters, reducing cognitive load and tool count.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | <=2025-11-25 | v2 |
Health check and diagnostic report for the Jambavan server and project configuration.
Search failure memory for past build errors, debugging sessions, and resolutions.
Record a build failure, error, or debugging session with symptom, attempted fix, and resolution.
Query the symbol dependency graph with actions: report (overview), query (search), or path (find call chains).
Analyze impact of changes on the codebase by tracing symbol dependencies.
Scan and index the project for symbols, functions, classes, and code structure.
Manage memory entries with modes: invalidate (soft-delete) or delete (permanent removal).
Extract and store decisions, TODOs, and facts from session text into project memory.
Read from project memory with actions: get (by id), search (by query), or recall (by scope).
Report on memory archive statistics and health.
Store a fact or decision in persistent project memory with title, body, and scope.
Generate a review pack with diffs, impact analysis, and context for code review.
Export or import session context and memories for handoff between agents or sessions.
Manage file watcher for incremental indexing with actions: start, stop, or status.
List files in a directory, respecting .gitignore and jambavan ignore rules.
Apply a text patch to a file by replacing old_text with new_text (requires JAMBAVAN_ALLOW_WRITE=1).
Read the contents of a file from the project.
Deprecated alias for jambavan_root_cause (counsel protocol).
Search for files and content matching a pattern in the project.
Deprecated alias for jambavan_strategy_plan (counsel protocol).
Deprecated alias for jambavan_verify_gate (counsel protocol).
Write content to a file in the project (requires JAMBAVAN_ALLOW_WRITE=1).
No error handling guidance visible. Tools like bash, write_file, and patch_file can fail in many ways (permission denied, file not found, syntax error) but no recovery hints or error classification (retryable vs. fatal) are documented. Agents cannot self-correct.
Parameter descriptions are sparse or missing for several tools. jambavan_graph 'symbol_limit' lacks explanation of what happens when limit is exceeded. jambavan_context 'compress_prose' and 'include_diff' lack detail on output impact. Agents cannot reason about parameter effects.