ContextStream MCP Integration

Persistent memory and cross-session learning for AI coding assistants. Cloud-based context management via MCP with semantic search, dependency graphs, and collaborative workflows.

What This Skill Does

This skill integrates ContextStream v0.4.x MCP server into your coding workflow, providing:

Step-by-Step Instructions

1. Session Initialization (First Message Only)

On the **first message** of every new session:

1. Call `session_init(folder_path="<cwd>", context_hint="<user_message>")`

2. Then call `context_smart(user_message="<user_message>", format="minified", max_tokens=400)`

3. If you see `[INGEST_RECOMMENDED]`, ask the user if they want to enable semantic code search

2. Every Subsequent Message

**BEFORE responding to any user message:**

1. Always call `context_smart(user_message="<user_message>", format="minified", max_tokens=400)` FIRST

2. This retrieves relevant context from previous sessions, decisions, and codebase

3. Use the returned context to inform your response

3. Search Protocol (CRITICAL)

**⚠️ STOP: Before using any local Search/Glob/Grep/Read/Explore tools:**

1. Call `search(mode="hybrid", query="...", limit=3)` FIRST

2. Only use local tools if ContextStream returns 0 results

3. Search modes:

- `semantic` - AI-powered conceptual search

- `hybrid` - Combines semantic + keyword

- `keyword` - Traditional text search

- `pattern` - Regex/glob patterns

4. Code Analysis

For understanding dependencies and impact:

5. Capturing Knowledge

**After completing work:**

```

session(action="capture", event_type="decision", title="Brief title", content="What was decided and why")

```

**When user corrects you or shows frustration:**

```

session(action="capture_lesson",

title="What went wrong",

trigger="What caused the mistake",

impact="How it affected the user",

prevention="How to avoid it next time")

```

**Before risky changes:**

```

session(action="get_lessons", query="<topic>")

```

6. Plans & Tasks

**Create a plan:**

```

session(action="capture_plan",

title="Plan Title",

description="...",

goals=["goal1", "goal2"],

steps=[{id: "1", title: "Step 1", order: 1}, ...])

```

**Create tasks:**

```

memory(action="create_task",

title="Task Title",

plan_id="<plan_id>",

priority="high|medium|low",

description="...")

```

**Update task status:**

```

memory(action="update_task",

task_id="<uuid>",

task_status="pending|in_progress|completed|blocked")

```

**List plans/tasks:**

7. Domain Tools Quick Reference

| Tool | Common Actions |

|------|----------------|

| `search` | `mode="semantic\|hybrid\|keyword\|pattern"` |

| `session` | `action="capture\|capture_lesson\|get_lessons\|recall\|smart_search"` |

| `memory` | `action="list_events\|create_task\|update_task\|list_tasks"` |

| `graph` | `action="dependencies\|impact\|call_path\|related\|ingest"` |

| `project` | `action="list\|overview\|index_status\|ingest_local"` |

| `workspace` | `action="list\|get\|associate"` |

| `integration` | `provider="github\|slack", action="search"` |

| `help` | `action="tools\|auth\|version\|editor_rules"` |

8. Context Pack (Pro+ Feature)

If Context Pack is enabled:

```

context_smart(user_message="...", format="minified", max_tokens=400, mode="pack", distill=true)

```

If unavailable, omit `mode` - the API will fall back automatically.

9. Tool Naming

ContextStream tools may be prefixed in your MCP client (e.g., `mcp__contextstream__search`). If a tool call fails with "No such tool available":

1. Call `help(action="tools")` to see exact tool names

2. Update your tool calls to match the exposed names

Important Constraints

Example Usage

**User:** "How does authentication work in this codebase?"

**Assistant actions:**

1. `context_smart(user_message="How does authentication work in this codebase?", format="minified", max_tokens=400)`

2. `search(mode="semantic", query="authentication login session", limit=5)`

3. `graph(action="dependencies", file_path="<auth_file>", symbol="authenticate")`

4. [Respond with findings]

5. `session(action="capture", event_type="decision", title="Authentication flow explained", content="User asked about auth, found in X, uses Y pattern")`

**User:** "Add OAuth support"

**Assistant actions:**

1. `context_smart(user_message="Add OAuth support", format="minified", max_tokens=400)`

2. `session(action="get_lessons", query="oauth authentication")`

3. `search(mode="hybrid", query="oauth config setup", limit=3)`

4. [Plan and implement]

5. `session(action="capture", event_type="decision", title="Added OAuth", content="Implemented OAuth using Z library, config in W")`

Additional Resources

Full documentation: https://contextstream.io/docs/mcp/tools