import type { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
const ID_NOTES = [
"Notes:",
"- If ContextStream is not initialized in this conversation, call `session_init` first (omit ids).",
"- Do not ask me for `workspace_id`/`project_id` — use session defaults or IDs returned by `session_init`.",
"- Prefer omitting IDs in tool calls when the tool supports defaults.",
];
const upgradeUrl = process.env.CONTEXTSTREAM_UPGRADE_URL || "https://contextstream.io/pricing";
const proPrompts = new Set<string>([
"build-context",
"generate-plan",
"generate-tasks",
"token-budget-context",
]);
function promptAccessLabel(promptName: string): "Free" | "PRO" {
return proPrompts.has(promptName) ? "PRO" : "Free";
}
export function registerPrompts(server: McpServer) {
server.registerPrompt(
"explore-codebase",
{
title: `Explore Codebase (${promptAccessLabel("explore-codebase")})`,
description: "Get an overview of a project codebase structure and key components",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"I want to understand the current codebase.",
"",
...ID_NOTES,
"",
"Please help me by:",
"1. Use `projects_overview` to get a project summary (use session defaults; only pass `project_id` if required).",
"2. Use `projects_files` to identify key entry points.",
"3. If a focus area is clear from our conversation, prioritize it; otherwise ask me what to focus on.",
"4. Use `search_semantic` (and optionally `search_hybrid`) to find the most relevant files.",
"5. Summarize the architecture, major modules, and where to start editing.",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"capture-decision",
{
title: `Capture Decision (${promptAccessLabel("capture-decision")})`,
description: "Document an architectural or technical decision in workspace memory",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Please capture an architectural/technical decision in ContextStream memory.",
"",
...ID_NOTES,
"",
"Instructions:",
"- If the decision is already described in this conversation, extract: title, context, decision, consequences/tradeoffs.",
"- If anything is missing, ask me 1–3 quick questions to fill the gaps.",
"- Then call `session_capture` with:",
' - event_type: "decision"',
" - title: (short ADR title)",
" - content: a well-formatted ADR (Context, Decision, Consequences)",
' - tags: include relevant tags (e.g., "adr", "architecture")',
' - importance: "high"',
"",
"After capturing, confirm what was saved.",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"review-context",
{
title: `Code Review Context (${promptAccessLabel("review-context")})`,
description: "Build context for reviewing code changes",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"I need context to review a set of code changes.",
"",
...ID_NOTES,
"",
"First:",
"- If file paths and a short change description are not already in this conversation, ask me for them.",
"",
"Then build review context by:",
"1. Using `graph_dependencies` to find what depends on the changed areas.",
"2. Using `graph_impact` to assess potential blast radius.",
"3. Using `memory_search` to find related decisions/notes.",
"4. Using `search_semantic` to find related code patterns.",
"",
"Provide:",
"- What the files/components do",
"- What might be affected",
"- Relevant prior decisions/lessons",
"- Review checklist + risks to focus on",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"investigate-bug",
{
title: `Investigate Bug (${promptAccessLabel("investigate-bug")})`,
description: "Build context for debugging an issue",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"I want help investigating a bug.",
"",
...ID_NOTES,
"",
"First:",
"- If the error/symptom is not already stated, ask me for the exact error message and what I expected vs what happened.",
"- If an affected area/component is not known, ask me where I noticed it.",
"",
"Then:",
"1. Use `search_semantic` to find code related to the error/symptom.",
"2. Use `search_pattern` to locate where similar errors are thrown or logged.",
"3. If you identify key functions, use `graph_call_path` to trace call flows.",
"4. Use `memory_search` to check if we have prior notes/bugs about this area.",
"",
"Return:",
"- Likely origin locations",
"- Call flow (if found)",
"- Related past context",
"- Suggested debugging steps",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"explore-knowledge",
{
title: `Explore Knowledge Graph (${promptAccessLabel("explore-knowledge")})`,
description: "Navigate and understand the knowledge graph for a workspace",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Help me explore the knowledge captured in this workspace.",
"",
...ID_NOTES,
"",
"Approach:",
"1. Use `memory_summary` for a high-level overview.",
"2. Use `memory_decisions` to see decision history (titles + a few key details).",
"3. Use `memory_list_nodes` to see available knowledge nodes.",
"4. If a starting topic is clear from the conversation, use `memory_search` for it.",
"5. Use `graph_related` on the most relevant nodes to expand connections.",
"",
"Provide:",
"- Key themes and topics",
"- Important decisions + rationale",
"- Suggested “next nodes” to explore",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"onboard-to-project",
{
title: `Project Onboarding (${promptAccessLabel("onboard-to-project")})`,
description: "Generate onboarding context for a new team member",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Create an onboarding guide for a new team member joining this project.",
"",
...ID_NOTES,
"",
"First:",
"- If the role is not specified, ask me what role they are onboarding into (backend, frontend, fullstack, etc.).",
"",
"Gather context:",
"1. Use `projects_overview` and `projects_statistics` for project summary.",
"2. Use `projects_files` to identify key entry points.",
"3. Use `memory_timeline` and `memory_decisions` to understand recent changes and architectural choices.",
"4. Use `search_semantic` to find READMEs/docs/setup instructions.",
"",
"Output:",
"- Project overview and purpose",
"- Tech stack + architecture map",
"- Key files/entry points relevant to the role",
"- Important decisions + rationale",
"- Recent changes/current focus",
"- Step-by-step getting started",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"analyze-refactoring",
{
title: `Refactoring Analysis (${promptAccessLabel("analyze-refactoring")})`,
description: "Analyze a codebase for refactoring opportunities",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Analyze the codebase for refactoring opportunities.",
"",
...ID_NOTES,
"",
"If a target area is obvious from our conversation, focus there; otherwise ask me what area to analyze.",
"",
"Please investigate:",
"1. `graph_circular_dependencies` (circular deps to break)",
"2. `graph_unused_code` (dead code to remove)",
"3. `search_pattern` (duplication patterns)",
"4. `projects_statistics` (high complexity hotspots)",
"",
"Provide a prioritized list with quick wins vs deeper refactors.",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"build-context",
{
title: `Build LLM Context (${promptAccessLabel("build-context")})`,
description: "Build comprehensive context for an LLM task",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Build comprehensive context for the task we are working on.",
"",
`Access: ${promptAccessLabel("build-context")}${promptAccessLabel("build-context") === "PRO" ? ` (upgrade: ${upgradeUrl})` : ""}`,
"",
...ID_NOTES,
"",
"First:",
"- If the “query” is clear from the latest user request, use that.",
"- Otherwise ask me: “What do you need context for?”",
"",
"Then:",
"- Call `ai_enhanced_context` with include_code=true, include_docs=true, include_memory=true (omit IDs unless required).",
"- Synthesize the returned context into a short briefing with links/file paths and key decisions/risks.",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"smart-search",
{
title: `Smart Search (${promptAccessLabel("smart-search")})`,
description: "Search across memory, decisions, and code for a query",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Find the most relevant context for what I am asking about.",
"",
...ID_NOTES,
"",
"First:",
"- If a query is clear from the conversation, use it.",
"- Otherwise ask me what I want to find.",
"",
"Then:",
"1. Use `session_smart_search` for the query.",
"2. If results are thin, follow up with `search_hybrid` and `memory_search`.",
"3. Return the top results with file paths/links and a short synthesis.",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"recall-context",
{
title: `Recall Context (${promptAccessLabel("recall-context")})`,
description: "Retrieve relevant past decisions and memory for a query",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Recall relevant past context (decisions, notes, lessons) for what I am asking about.",
"",
...ID_NOTES,
"",
"First:",
"- If a recall query is clear from the conversation, use it.",
"- Otherwise ask me what topic I want to recall.",
"",
"Then:",
"- Use `session_recall` with the query (omit IDs unless required).",
"- Summarize the key points and any relevant decisions/lessons.",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"session-summary",
{
title: `Session Summary (${promptAccessLabel("session-summary")})`,
description: "Get a compact summary of workspace/project context",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Generate a compact, token-efficient summary of the current workspace/project context.",
"",
...ID_NOTES,
"",
"Use `session_summary` (default max_tokens=500 unless I specify otherwise).",
"Then list:",
"- Top decisions (titles only)",
"- Any high-priority lessons to watch for",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"capture-lesson",
{
title: `Capture Lesson (${promptAccessLabel("capture-lesson")})`,
description: "Record a lesson learned from an error or correction",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Capture a lesson learned so it is surfaced in future sessions.",
"",
...ID_NOTES,
"",
"If the lesson details are not fully present in the conversation, ask me for:",
"- title (what to remember)",
"- severity (low|medium|high|critical, default medium)",
"- category (workflow|code_quality|verification|communication|project_specific)",
"- trigger (what caused it)",
"- impact (what went wrong)",
"- prevention (how to prevent it)",
"- keywords (optional)",
"",
"Then call `session_capture_lesson` with those fields and confirm it was saved.",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"capture-preference",
{
title: `Capture Preference (${promptAccessLabel("capture-preference")})`,
description: "Save a user preference to memory",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Save a user preference to ContextStream memory.",
"",
...ID_NOTES,
"",
"If the preference is not explicit in the conversation, ask me what to remember.",
"",
"Then call `session_capture` with:",
'- event_type: "preference"',
"- title: (short title)",
"- content: (preference text)",
'- importance: "medium"',
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"capture-task",
{
title: `Capture Task (${promptAccessLabel("capture-task")})`,
description: "Capture an action item into memory",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Capture an action item into ContextStream memory.",
"",
...ID_NOTES,
"",
"If the task is not explicit in the conversation, ask me what to capture.",
"",
"Then call `session_capture` with:",
'- event_type: "task"',
"- title: (short title)",
"- content: (task details)",
'- importance: "medium"',
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"capture-bug",
{
title: `Capture Bug (${promptAccessLabel("capture-bug")})`,
description: "Capture a bug report into workspace memory",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Capture a bug report in ContextStream memory.",
"",
...ID_NOTES,
"",
"If details are missing, ask me for:",
"- title",
"- description",
"- steps to reproduce (optional)",
"- expected behavior (optional)",
"- actual behavior (optional)",
"",
"Then call `session_capture` with:",
'- event_type: "bug"',
"- title: (bug title)",
"- content: a well-formatted bug report (include all provided details)",
"- tags: component/area tags",
'- importance: "high"',
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"capture-feature",
{
title: `Capture Feature (${promptAccessLabel("capture-feature")})`,
description: "Capture a feature request into workspace memory",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Capture a feature request in ContextStream memory.",
"",
...ID_NOTES,
"",
"If details are missing, ask me for:",
"- title",
"- description",
"- rationale (optional)",
"- acceptance criteria (optional)",
"",
"Then call `session_capture` with:",
'- event_type: "feature"',
"- title: (feature title)",
"- content: a well-formatted feature request",
"- tags: component/area tags",
'- importance: "medium"',
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"generate-plan",
{
title: `Generate Plan (${promptAccessLabel("generate-plan")})`,
description: "Generate a development plan from a description",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Generate a development plan for what I am trying to build/fix.",
"",
`Access: ${promptAccessLabel("generate-plan")}${promptAccessLabel("generate-plan") === "PRO" ? ` (upgrade: ${upgradeUrl})` : ""}`,
"",
...ID_NOTES,
"",
"Use the most recent user request as the plan description. If unclear, ask me for a one-paragraph description.",
"",
"Then call `ai_plan` and present the plan as an ordered list with milestones and risks.",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"generate-tasks",
{
title: `Generate Tasks (${promptAccessLabel("generate-tasks")})`,
description: "Generate actionable tasks from a plan or description",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Generate actionable tasks for the work we are discussing.",
"",
`Access: ${promptAccessLabel("generate-tasks")}${promptAccessLabel("generate-tasks") === "PRO" ? ` (upgrade: ${upgradeUrl})` : ""}`,
"",
...ID_NOTES,
"",
"If a plan_id exists in the conversation, use it. Otherwise use the latest user request as the description.",
"If granularity is not specified, default to medium.",
"",
"Call `ai_tasks` and return a checklist of tasks with acceptance criteria for each.",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"token-budget-context",
{
title: `Token-Budget Context (${promptAccessLabel("token-budget-context")})`,
description: "Get the most relevant context that fits within a token budget",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Build the most relevant context that fits within a token budget.",
"",
`Access: ${promptAccessLabel("token-budget-context")}${promptAccessLabel("token-budget-context") === "PRO" ? ` (upgrade: ${upgradeUrl})` : ""}`,
"",
...ID_NOTES,
"",
"First:",
"- If a query is clear from the conversation, use it; otherwise ask me for a query.",
"- If max_tokens is not specified, ask me for a token budget (e.g., 500/1000/2000).",
"",
"Then call `ai_context_budget` and return the packed context plus a short note about what was included/excluded.",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"find-todos",
{
title: `Find TODOs (${promptAccessLabel("find-todos")})`,
description: "Scan the codebase for TODO/FIXME/HACK notes and summarize",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Scan the codebase for TODO/FIXME/HACK notes and summarize them.",
"",
...ID_NOTES,
"",
"Use `search_pattern` with query `TODO|FIXME|HACK` (or a pattern inferred from the conversation).",
"Group results by file path, summarize themes, and propose a small prioritized cleanup list.",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"generate-editor-rules",
{
title: `Generate Editor Rules (${promptAccessLabel("generate-editor-rules")})`,
description: "Generate ContextStream AI rule files for your editor",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Generate ContextStream AI rule files for my editor.",
"",
...ID_NOTES,
"",
"First:",
"- If you can infer the project folder path from the environment/IDE roots, use it.",
"- Otherwise ask me for an absolute folder path.",
"- Ask which editor(s) (codex,cursor,cline,kilo,roo,claude,aider,antigravity) or default to all.",
"",
"Then call `generate_rules` and confirm which files were created/updated.",
"Ask if the user also wants to apply rules globally (pass apply_global: true).",
].join("\n"),
},
},
],
})
);
server.registerPrompt(
"index-local-repo",
{
title: `Index Local Repo (${promptAccessLabel("index-local-repo")})`,
description: "Ingest local files into ContextStream for indexing/search",
},
async () => ({
messages: [
{
role: "user",
content: {
type: "text",
text: [
"Ingest local files into ContextStream for indexing/search.",
"",
...ID_NOTES,
"",
"First:",
"- Ask me for the local directory path to ingest if it is not already specified.",
"",
"Then:",
"- Call `projects_ingest_local` with the path (use session defaults for project, or the `project_id` returned by `session_init`).",
"- Explain how to monitor progress via `projects_index_status`.",
].join("\n"),
},
},
],
})
);
}
