Skip to main content
Glama
Karim-Os

mcp-pricescout

by Karim-Os
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Remote

In this project, you are going to make a chatbot to scrape LLM Inference Serving websites to research costs of serving various LLMs. You will do this by writing an MCP Server that hooks up to Firecrawl's API and saving the data in a SQLite Database. You should use the following websites to scrape:

  1. Make a venv with uv

  2. Sync venv with pyproject.toml (uv sync)

  3. Make an API Key on Anthropic and Firecrawl

  4. Complete the 2 tool calls in starter_server.py

  5. Change the server_config.json to point to your server file

  6. Complete any section in starter_client.py that has "#complete".

  7. Test using any methods taught in the course

  8. Use the following prompts in your chatbot but play around with all the LLM providers in the list above:

Environment notes

Two dependency-drift issues surfaced against the packages actually resolved at implementation time (Aug 2026), fixed rather than worked around:

  • firecrawl-py's .scrape() has no success key. The currently-installed firecrawl-py (v4.38.0, the only version whose .scrape(url, formats=...) method matches this assignment's shape) raises on failure and returns a Document with no success field on success — unlike the older API this assignment's instructions describe. starter_server.py's scrape_result.get("success", True) defaults to True instead of False so real successful scrapes are actually detected (an explicit success: false, if a future/older version ever returns one, still works).

  • mcp-server-sqlite crashes against the latest mcp package. Its last release (Apr 2025) pins mcp[cli]>=1.6.0 with no upper bound, and current mcp releases removed the Server.list_resources decorator it depends on. server_config.json's sqlite entry pins a compatible mcp version for that one subprocess only, via uvx --with "mcp<1.10" mcp-server-sqlite ... — this doesn't affect the mcp version this project itself uses.

If using a Vocareum-proxied Anthropic key (voc-...), set ANTHROPIC_BASE_URL=https://claude.vocareum.com in .env alongside ANTHROPIC_API_KEY — starter_client.py reads it and passes it to the Anthropic client; leaving it unset uses the standard Anthropic API.

Related MCP server: LLM Pricing Scraper MCP Server

Verified live

Real end-to-end run, evidence in evidence/ — full transcript, real scraped content, and what each piece of it proves. All four required items are demonstrated, including a real Firecrawl scrape (Successfully scraped N out of N websites). Three real bugs surfaced only by actually running this, none visible in code review or offline testing:

  • load_dotenv() doesn't override an already-set environment variable by default. This machine has an ambient user-level ANTHROPIC_BASE_URL pointing at the standard API (unrelated to this project), which silently took precedence over .env's Vocareum URL — every request went to api.anthropic.com with a voc-... key and got a clean 401 invalid x-api-key. Fixed with load_dotenv(override=True), which is the correct behavior for a project's own .env regardless of ambient state.

  • process_query's tool-use loop broke when a single response contained more than one tool_use block — a real, common pattern (Claude often batches several tool calls in one turn). The original loop called self.anthropic.messages.create() again inside the for content in response.content: loop, reassigning response while that same for loop kept iterating the old content list — so a second tool_use in the same response never got its matching tool_result appended, and the next API call failed with 400: tool_use ids were found without tool_result blocks immediately after. Fixed by collecting every block of a response (text to print, all tool_use calls to execute) before making the next call, and sending all of that turn's tool_results together in one message, matching what the API actually requires.

  • mcp-server-sqlite's read_query returns Python repr()-style output, not JSON — single-quoted ("[{'company_name': 'x', ...}]"), which looks like JSON at a glance but makes json.loads() fail immediately (Expecting property name enclosed in double quotes). show_stored_data now parses it with ast.literal_eval() instead, which handles the format actually being sent.

Available Tools

2 tools
extract_scraped_infoB

Extract information about a scraped website.

Args: identifier: The provider name, full URL, or domain to look for

Returns: Formatted JSON string with the scraped information

ParametersJSON Schema
NameRequiredDescriptionDefault
identifierYes

Output Schema

ParametersJSON Schema
NameRequiredDescription
resultYes

TDQS

B3/5.0
Behavior2/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

No annotations are provided, so the description carries the full burden. It only says 'extract' and that it returns a JSON string, but does not disclose whether this is a read-only operation, what happens if the identifier is not found, or any rate limits or error conditions. Minimal disclosure of behavior beyond the basic action.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is very short and to the point, with the main purpose front-loaded. No redundant phrasing or filler. It is concise without being under-specified.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness2/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a simple one-parameter tool with an output schema, the description is minimal. It lacks context about the relationship with scrape_websites, any prerequisites, or potential failure modes. The output schema covers the return format, but the description does not tell the agent enough about when or why to use this tool.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 0%, but the description compensates by explaining that 'identifier' can be a provider name, full URL, or domain. This adds meaningful semantics beyond the schema's bare 'Identifier' label, though it could be more specific about accepted formats.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose4/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a clear verb (extract) and resource (scraped website), and the sibling scrape_websites is clearly different (scraping vs extracting). It does not specify what 'information' includes, but the purpose is unambiguous.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines2/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

No guidance on when to use this tool versus the sibling scrape_websites. The description does not mention prerequisites (e.g., that the website must have been scraped first) or any conditions that would make this tool the right choice.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

scrape_websitesC

Scrape multiple websites using Firecrawl and store their content.

Args: websites: Dictionary of provider_name -> URL mappings formats: List of formats to scrape ['markdown', 'html'] (default: both) api_key: Firecrawl API key (if None, expects environment variable)

Returns: List of provider names for successfully scraped websites

ParametersJSON Schema
NameRequiredDescriptionDefault
api_keyNo
formatsNo
websitesYes

Output Schema

ParametersJSON Schema
NameRequiredDescription
resultYes

TDQS

C2.5/5.0
Behavior1/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations, the description carries the full burden of disclosing side effects, limitations, or persistence. It only mentions that content is 'stored,' but does not elaborate on any I/O implications, error handling, or whether the operation is read-only. This is insufficient for an agent to anticipate the tool's behavior.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is concise and to the point, using a single sentence to convey the primary action and then listing parameters without unnecessary verbosity. It is easily scannable and does not waste words, though it could be slightly richer without losing brevity.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness2/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the complexity (nested objects, multiple parameters, and an output schema), the description is incomplete. It does not explain the semantics of the returned list of provider names, potential failure cases, or how the output relates to the input. The presence of a sibling tool also suggests a broader workflow that is not addressed, leaving the agent with gaps in understanding the tool's place in a larger process.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters2/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The description adds minimal meaning beyond the schema. It restates that 'websites' is a dictionary of provider_name to URL, but does not explain the purpose of provider_name, the valid values for 'formats' (beyond examples), or the fallback behavior when 'api_key' is None. Since schema coverage is 0%, the description should have compensated but does not.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose4/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the action (scrape multiple websites) and the resource (websites), and mentions the use of Firecrawl and content storage. It is specific enough to understand the tool's primary function, though it could be clearer about how it differs from the sibling tool 'extract_scraped_info'.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines2/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description provides no guidance on when to use this tool versus the sibling tool 'extract_scraped_info'. It does not mention that this tool is for raw scraping while the sibling is for extracting structured information, leaving the agent to infer the appropriate context without explicit instructions.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

  1. 2 tool updatesv0.1.0
    • First observedextract_scraped_info
    • First observedscrape_websites
Deploy Server
One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

TDQS

B3.3/5.0

Scored across 2 tools

Disambiguation4/5

The two tools have clearly distinct roles: one initiates the scraping workflow and the other retrieves previously scraped data. There is minor potential for confusion about what 'extract' means in the naming, but the descriptions resolve the boundary.

Naming Consistency5/5

Both tools follow a consistent verb_object pattern with snake_case: scrape_websites and extract_scraped_info. The naming style is uniform and predictable.

Tool Count3/5

Two tools is on the low end for a price-scouting server, but the pair covers a minimal scrape-then-retrieve workflow. It feels thin rather than bloated, so it is borderline acceptable.

Completeness2/5

The server appears aimed at price monitoring, but there is no tool for searching, comparing, updating, or deleting scraped data. The surface only supports scraping and retrieving stored content, leaving significant gaps for a typical price-scout use case.

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • mcp-price-scout

    lysss20
    F
    license
    Not graded
    quality
    C
    maintenance
    Enables scraping LLM inference pricing from websites like Cloudrift, DeepInfra, Fireworks, and Groq, storing results in SQLite for cost comparison.
    -