Langsmith Mcp Server

Fetch prompts from LangSmith with optional filtering. Args: is_public (str): Filter by prompt visibility - "true" for public prompts, "false" for private prompts (default: "false") limit (int): Maximum number of prompts to return (default: 20) Returns: Dict[str, Any]: Dictionary containing the prompts and metadata

Get a specific prompt by its exact name. Args: prompt_name (str): The exact name of the prompt to retrieve ctx: FastMCP context (automatically provided) Returns: Dict[str, Any]: Dictionary containing the prompt details and template, or an error message if the prompt cannot be found

Call this tool when you need to understand how to create and push prompts to LangSmith.

Retrieve one page of message history for a specific conversation thread. Uses char-based pagination: pages are built by character budget (max_chars_per_page). Long strings are truncated to preview_chars. Supply page_number (1-based) on every call; use the returned total_pages to request further pages. Args: thread_id (str): The unique ID of the thread to fetch history for project_name (str): The name of the project containing the thread (format: "owner/project" or just "project") page_number (int): 1-based page index (required) max_chars_per_page (int): Max character count per page, capped at 30000 (default: 25000) preview_chars (int): Truncate long strings to this length with "… (+N chars)" (default: 150) Returns: Dict with result (list of messages), page_number, total_pages, max_chars_per_page, preview_chars; or an error message if the thread cannot be found

Fetch LangSmith runs from one or more projects with flexible filters and automatic pagination. All results are paginated by character budget to keep responses manageable. Use page_number and total_pages from the response to iterate through multiple pages. --- 📤 RETURNS (always paginated) ------------------------------ Dict with: runs, page_number, total_pages, max_chars_per_page, preview_chars. May include _truncated, _truncated_message, _truncated_preview if content exceeded budget. --- ⚙️ PARAMETERS (required) ------------------------ project_name : str The project name to fetch runs from. For multiple projects, use JSON array string: '["project1", "project2"]' limit : int (required) Maximum number of runs to fetch from LangSmith API (capped at 100). These runs are then paginated by character budget into pages. page_number : int, default 1 1-based page index. Use with total_pages from response to iterate through pages. trace_id : str, optional Return only runs that belong to this trace UUID. Example: "123e4567-e89b-12d3-a456-426614174000" run_type : str, optional Filter runs by type: "llm", "chain", "tool", "retriever". error : str, optional "true" for errored runs, "false" for successful runs. is_root : str, optional "true" for only top-level traces, "false" to exclude roots. filter : str, optional Filter Query Language (FQL) expression. Common fields: id, name, run_type, start_time, end_time, latency, total_tokens, error, tags, feedback_key, feedback_score, metadata_key, metadata_value, execution_order. Operators: eq, neq, gt, gte, lt, lte, has, search, and, or, not. Examples: 'gt(latency, "5s")' # runs > 5 seconds 'neq(error, null)' # errored runs 'has(tags, "beta")' # tagged "beta" 'and(eq(name,"ChatOpenAI"), eq(run_type,"llm"))' # name AND type 'search("image classification")' # full-text search trace_filter : str, optional Filter on the root run of each trace tree. Example: 'and(eq(feedback_key,"user_score"), eq(feedback_score,1))' tree_filter : str, optional Filter on any run in the trace tree (siblings/children included). Example: 'eq(name,"ExpandQuery")' order_by : str, default "-start_time" Sort field; prefix "-" for descending. Examples: "-start_time", "latency" reference_example_id : str, optional Filter runs by dataset example ID. max_chars_per_page : int, default 25000 Max JSON character count per page (capped at 30000). Pagination splits by this budget. preview_chars : int, default 150 Truncate long strings to this length with "… (+N chars)". Keeps responses readable. --- 🧪 EXAMPLES ----------- 1️⃣ Get latest 10 root runs: fetch_runs("my-project", limit=10, page_number=1, is_root="true") 2️⃣ Get errored tool runs: fetch_runs("my-project", limit=50, page_number=1, run_type="tool", error="true") 3️⃣ Runs > 5s with "experimental" tag: fetch_runs("my-project", limit=50, page_number=1, filter='and(gt(latency,"5s"), has(tags,"experimental"))') 4️⃣ All runs for a trace (paginate through pages): r = fetch_runs("my-project", limit=50, page_number=1, trace_id="abc-123") # Check r["total_pages"] and fetch subsequent pages if needed